ThinkMorph/ThinkMorph-SSL-Instruct
Any-to-Any • Updated
id stringlengths 64 64 | subset stringclasses 20
values | source_split stringclasses 2
values | question stringlengths 24 1.09k | options stringlengths 0 642 ⌀ | answer stringclasses 19
values | input_images images listlengths 1 4 | reasoning_thought_0 stringlengths 63 3.99k | reasoning_image_0 imagewidth (px) 288 512 | reasoning_thought_1 stringlengths 0 3.2k | reasoning_image_1 imagewidth (px) 288 512 ⌀ | reasoning_thought_2 stringlengths 0 726 | reasoning_image_2 imagewidth (px) 512 512 ⌀ | reasoning_thought_3 stringclasses 11
values |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
01726f7b8af34fff292afd73ba794abf1bce5255100fec3f03d99bb12c1f41d8 | spatial/behind_me_1k | train | Based on these four images (image 1, 2, 3, and 4) showing the pink ball from different viewpoints, with each camera aligned with room walls and partially capturing the surroundings: If I am standing at the same spot and facing the same direction as shown in image 3, what is behind me? A. Cardboard-covered sliding door ... | null | B | In the queried view, the camera is facing the cardboard-covered sliding door. To determine what is behind that viewpoint, I first need a bird's-eye view centered on the pink ball so the arrangement of the nearby reference objects on the bed becomes stable and the facing direction in the queried scene can be tracked wit... | The first bird's-eye view shows the pink ball with the purple cloth, patterned cloth, white iron, green device, and red-white box laid out clearly around it. Matching those bed items to the queried scene fixes the current orientation: from image 3, the camera is looking toward the side of the bed where the iron appears... | The transformed bird's-eye view is the turned-around orientation, and it matches the arrangement seen in the view with the vent hood and flexible duct above the bed. That means the direction in front of that matching original view is exactly what lies behind the observer in image 3. In that matched room view, the promi... | Not supported with pagination yet | ||||
235cffbb23b7c1919924c4d56386cffe3014d4505ae5257783027d7d365163c9 | spatial/behind_me_1k | train | Based on these four images (image 1, 2, 3, and 4) showing the bottle from different viewpoints, with each camera aligned with room walls and partially capturing the surroundings: If I am standing at the same spot and facing the same direction as shown in image 2, what is behind me? A. Gate B. Stone fountain C. Light br... | null | A | In the queried view, the bottle is centered with the incense burner slightly to its left, and the background is split between a plain light-brown wall on one side and a decorated lattice-style wall on the other. To answer what is behind the viewer in that stance, I first need a bird's-eye view of this arrangement so th... | The bird's-eye view shows the bottle below the incense burner, with the incense burner farther ahead in the viewing direction and the metal cup off toward the lower-left area of the table. Since 'behind me' means staying at the same spot but turning 180 degrees, I next rotate this top-down layout to the opposite facing... | The transformed bird's-eye view matches the orientation seen in image 4: after the turn, the incense burner is near the foreground, the bottle remains central, and the metal cup appears toward the right side. Looking at image 4, the scene in front is the glass entrance/gate area. Therefore, what is behind the viewer in... | Not supported with pagination yet | ||||
85c01ba1d9bea9da986ece04a43d9a8a1cec8ba8793980939fa528fcdacf35a1 | spatial/behind_me_1k | train | Based on these four images (image 1, 2, 3, and 4) showing the green ball from different viewpoints, with each camera aligned with room walls and partially capturing the surroundings: If I am standing at the same spot and facing the same direction as shown in image 4, what is behind me? A. Window B. Cardboard-covered sl... | null | A | In the queried view, image 4, the camera faces a plain tiled wall, with the green ball near the middle of the bed and the white mouse to its left. The other views show the same bed from other sides, with distinctive landmarks like a window, a cardboard-covered sliding door, and equipment on another wall. To answer what... | The bird's-eye view of the queried viewpoint makes the bed layout easy to compare: the mouse sits to the left of the ball, the purple cloth is toward the lower-left area, and the patterned cloth is lower-middle, matching image 4's orientation. Since the question asks what is behind me, the needed orientation change is ... | The transformed bird's-eye view matches the arrangement seen in image 2: after the turn, the mouse is on the right of the ball, the purple cloth is on the upper-right side, and the small green box is toward the upper-left. So the opposite direction from image 4 corresponds to the viewpoint in image 2. Looking at that o... | Not supported with pagination yet | ||||
80fb91124a0d5a6ba04c1f9448392e8a5e7a999907c615fd05c1723f3afb6e1d | spatial/behind_me_1k | train | Based on these four images (image 1, 2, 3, and 4) showing the green toy fighter from different viewpoints, with each camera aligned with room walls and partially capturing the surroundings: If I am standing at the same spot and facing the same direction as shown in image 1, what is behind me? A. Wall B. Window C. Lots ... | null | B | In the queried view, the toy fighter sits on the round table with a plain tiled wall ahead. The other views show the same table from different sides: one side has the floral folding screen, another has the sink and window, and another has a pile of toys against a wall. To answer what is behind the viewer in the queried... | The first generated bird's-eye view stabilizes the table layout: the watermelon-pattern cloth is on one side, the soap-like white object is off to one side, and the blue tube is on another. This lets me anchor the queried viewpoint relative to the table. Since the question asks what is behind me, the needed orientation... | The transformed bird's-eye view matches the orientation seen in the window-side original photo: the blue tube and white object swap to the corresponding sides exactly as they do when facing the sink and window. That means the direction in front of that matched original view is the direction behind the viewer in the que... | Not supported with pagination yet | ||||
e898e79df446e42a3d5bfcacdeae72f44bd1593f30a5f5c55e7b1558ac8e2787 | spatial/behind_me_1k | val | Based on these four images (image 1, 2, 3, and 4) showing the blue ball from different viewpoints, with each camera aligned with room walls and partially capturing the surroundings: If I am standing at the same spot and facing the same direction as shown in image 4, what is behind me? A. School bag and TV cabinet B. Da... | null | B | From the queried view in image 4, the blue ball sits on the same round covered surface seen in the other views, so the ball can be used as the common reference point for camera position. To answer what is behind the viewer in image 4, I first need a bird's-eye view of that queried viewpoint so the facing direction can ... | The first generated bird's-eye view fixes the round surface and ball from above, giving a clean orientation reference for the image-4 viewpoint. Since the question asks what is behind me, the needed change is not moving to a new place but rotating 180 degrees at the same spot. I therefore generate a second top-down vie... | The transformed bird's-eye view matches the orientation seen in image 2. Looking back to that original room view, the object directly ahead is the dark brown sofa. Because this transformed front direction is exactly what lies behind the viewer in image 4, the answer is B. Dark brown sofa. | Not supported with pagination yet | ||||
f416e2b1b71a0d619d7b8d2bdfa1567c7e2a20803d687c12337eff3944b8c0e8 | spatial/behind_variants_286 | val | Based on these three images (image 1, 2, and 3) showing the same scene from different viewpoints: From the viewpoint presented in image 2, what is behind of the stone elephant sculpture. A. ceramics planter B. red wall C. lighter | null | A | In the queried viewpoint, the stone elephant is seen from above and slightly from its left side, with the planter partly occluded near the elephant’s far side. To determine what is behind the elephant from that exact viewing direction, I will generate a bird's-eye view aligned with the same orientation as the queried v... | The generated bird's-eye view shows the elephant’s body below and the ceramics planter positioned directly beyond it in the behind direction relative to the queried viewpoint. The red wall runs along the left edge of the scene rather than directly behind the sculpture, and the small ground light is off to the side near... | Not supported with pagination yet | Not supported with pagination yet | ||||
62d70cfba9a1cda21a9f107272219e3b073bc3e65c63a96d89248dd51fa091f1 | spatial/behind_variants_286 | val | Based on these three images (image 1, 2, and 3) showing the same scene from different viewpoints: From the viewpoint presented in image 2, what is the nearest object the nearest object behind of the umbrella cover dispenser. A. chair B. trash bin C. glass door D. warning sign | null | A | The question asks about what is behind the umbrella cover dispenser specifically from the viewpoint of the side-on scene in image 2. To judge "behind" correctly, I should generate a bird's-eye view aligned with that same viewing direction, because a top-down layout makes the front/behind relationship around the dispens... | In the generated bird's-eye view, the umbrella cover dispenser sits near the lower-left beside the glass wall, and the two-seat chair is immediately to its right. Because this top-down layout is aligned with the viewpoint of image 2, the region behind the dispenser is the area just past it toward the chair side. The ne... | Not supported with pagination yet | Not supported with pagination yet | ||||
89945cd0fbb87a7b1875012c8452a0e6379cf4bd71cba817bc621c97440508b6 | spatial/behind_variants_286 | val | Based on these three images (image 1, 2, and 3) showing the same scene from different viewpoints: From the viewpoint presented in image 2, is there a "7-ELEVEN" cart behind the planter box? A. No. B. Yes | null | B | In the queried viewpoint, the planter box fills the foreground and blocks part of the area directly beyond it, so the key is to determine what lies behind that planter from the same viewing direction as image 2. To make the behind relationship unambiguous, I will generate a bird's-eye view aligned with that viewpoint, ... | The generated bird's-eye view shows the planter box running vertically in the lower-middle area, with the viewing direction matching image 2. In this aligned top-down layout, the space behind the planter is the area farther away from the viewer, above it in the scene. There, a rectangular cart is clearly positioned jus... | Not supported with pagination yet | Not supported with pagination yet | ||||
756b3b6841b7cb8043b0187006d8a92f893a9a7c6426d6f20baf7da96d3dd13b | spatial/behind_variants_286 | val | Based on these three images (image 1, 2, and 3) showing the same scene from different viewpoints: From the viewpoint presented in image 3, what is the nearest object behind of the light-green electrical box. A. dark-green electrical box B. car C. bike D. greenery | null | A | In the queried street-side view, the light-green electrical box is the open central cabinet, and the question asks what is behind it from the same viewpoint as that image. To judge 'behind' reliably, I will generate a bird's-eye view aligned with that starting viewpoint, because a top-down layout makes the front/behind... | The generated bird's-eye view shows the same arrangement from above with the queried viewpoint orientation preserved: the curb and greenery are along the bottom, the road is toward the top, and the row of utility boxes runs left-to-right. The light-green electrical box is the narrow central box between the tall light-g... | Not supported with pagination yet | Not supported with pagination yet | ||||
c2fc283f042c36d86a1eeaa8bce543a9c9fe8e4b60337e7d56d2425a28a417ca | spatial/behind_variants_286 | val | Based on these three images (image 1, 2, and 3) showing the same scene from different viewpoints: From the viewpoint presented in image 3, what is behind of the fire hydrant. A. bollard light B. gutter C. road | null | A | In the queried view, the red fire hydrant box is in the foreground and the gray bollard light appears farther back, but perspective alone can make 'behind' ambiguous. To resolve this cleanly, I would generate a bird's-eye view aligned with the same facing direction as the queried viewpoint, because a top-down layout ma... | The generated bird's-eye view shows the same scene from above while preserving the queried viewpoint's orientation. In that layout, the red fire hydrant sits closer to the bottom, and the region behind it from the image-3 viewpoint extends upward. In that behind region, the gray bollard light is positioned directly bey... | Not supported with pagination yet | Not supported with pagination yet | ||||
aa45b3b692ee0e68c1c21b465859c1597edc802997818e57686a3c6517cd72af | spatial/directional_query_2k | train | Based on these four images (image 1, 2, 3, and 4) showing the powder bottle from different viewpoints, with each camera aligned with room walls and partially capturing the surroundings: From the viewpoint presented in image 3, what is to the left of the powder bottle? A. Dining area B. Black cabinet C. Red bed frame wi... | null | A | I first inspect the queried view, the one with the red bed frame behind the powder bottle. The question asks what is to the left of the bottle from that specific viewpoint, so I need to convert this egocentric left/right relation into the room layout. To do that, I will generate a bird's-eye view of the tabletop and ne... | The first generated bird's-eye view shows the powder bottle near the center-right of the blue mat, with the black round container and green packet on one side, and the tissue pack and green rectangular pouch on the other. This top-down layout clarifies where objects sit around the bottle independent of perspective dist... | In the rotated bird's-eye view, the bottle's aligned forward direction matches the original viewpoint that looks out toward the open room with the table and chairs in the distance. That is the same orientation as the image showing the dining area beyond the tabletop. In that matched original view, the space directly ah... | Not supported with pagination yet | ||||
f5f16d95157c9f7053aec420b1e63cfce8a4399e1d08dc9a8c94a278a39157d7 | spatial/directional_query_2k | train | Based on these four images (image 1, 2, 3, and 4) showing the yellow bottle from different viewpoints, with each camera aligned with room walls and partially capturing the surroundings: From the viewpoint presented in image 1, what is to the right of the yellow bottle? A. Wall B. Black chairs C. Glass wall and black ch... | null | B | In the queried view, the yellow bottle is centered on the table, and the background shows an office area with dark monitors and several black chairs. To answer what is to the bottle’s right from that exact viewpoint, I first want to form a bird’s-eye view of the table and bottle. That top-down layout helps convert the ... | The first generated bird’s-eye view shows the bottle near the middle of the tabletop and fixes the queried camera orientation relative to the table pattern. Since the question asks for the bottle’s right side in the original perspective, I next rotate this top-down layout by 90 degrees anticlockwise so that the bottle’... | The rotated top-down view aligns with the viewpoint seen in the original image where the bottle faces toward the office area with the black seating visible behind it. Using that aligned original perspective, the area in front of the bottle corresponds to the right side of the bottle in the queried view. In that aligned... | Not supported with pagination yet | ||||
35f9c9655a0b54e5cf94a34ab08c4ae918547ba07bf218d559dc6d9689762d34 | spatial/directional_query_2k | train | Based on these four images (image 1, 2, 3, and 4) showing the green ball from different viewpoints, with each camera aligned with room walls and partially capturing the surroundings: From the viewpoint presented in image 1, what is to the left of the green ball? A. White-red cabinet B. School bag and TV cabinet C. Ligh... | null | A | In the queried view, the green ball is seen with the bed and beige backrest behind it, but a left/right judgment from this camera angle is easier if I first convert the scene into a bird's-eye layout centered on the ball. I will generate a top-down view of the queried setup so I can map the camera-facing left side of t... | The bird's-eye view shows the round covered surface clearly, with the green ball slightly off center and the surrounding cloth patterns giving a stable orientation for the queried viewpoint. From this overhead layout, the asked direction is the ball's left as seen from image 1, so I need to rotate the top-down layout b... | In the rotated top-down view, the cloth pattern orientation now matches the viewpoint seen in the fourth original image, so that image is the normal-perspective view looking toward the side that is left of the ball in image 1. In that aligned original view, the prominent object beyond the ball is the white cabinet with... | Not supported with pagination yet | ||||
6cd0b3e9cad34b27c2221f8a88080598c8b36e877545695ac5a011a85b9481c8 | spatial/directional_query_2k | train | Based on these four images (image 1, 2, 3, and 4) showing the black sneaker from different viewpoints, with each camera aligned with room walls and partially capturing the surroundings: From the viewpoint presented in image 4, what is to the right of the black sneaker? A. Washing machine B. Wall and window C. Curtain D... | null | D | I first inspect the queried perspective, which is the front-facing view of the black sneaker in image 4, and compare it with the other views to understand the room layout around the table. Because the question asks what is to the right of the sneaker from that specific viewpoint, I want to first generate a bird's-eye v... | The first generated bird's-eye view shows the sneaker centered on the table with surrounding items clearly arranged around it: a green bottle and small red object on one side, a spray can toward another edge, and clips and papers on the opposite side. This overhead view clarifies the local object positions, but to answ... | In the rotated top-down view, the orientation now matches one of the original camera directions: it aligns with the side view seen in image 3, where the sneaker points leftward across the frame. In that aligned original view, the area directly in front of the sneaker corresponds to what was to the right of the sneaker ... | Not supported with pagination yet | ||||
6d2ebd395e73b8a9c56b3dbda3c9bd3d97b71220055e8df663323e4ce99ddd72 | spatial/directional_query_2k | train | Based on these four images (image 1, 2, 3, and 4) showing the pepsi bottle from different viewpoints, with each camera aligned with room walls and partially capturing the surroundings: From the viewpoint presented in image 3, what is to the right of the pepsi bottle? A. Light-colored sofa B. Red stool C. Cabinet and po... | null | A | I first inspect the queried view, which is the third photo. The pepsi bottle faces a hallway-like space with a cabinet and potted plant behind it, while other photos show different room sides: one with a dark brown sofa, one with a red stool and low cabinet, and one with a light-colored sofa. To answer what is to the r... | The first generated bird's-eye view shows the round tablecloth pattern around the bottle from above, letting me treat the bottle as the center of a compass. For a right-side query, I rotate this top-down layout 90 degrees anticlockwise so that whatever was to the right in the third photo becomes the forward direction i... | In the rotated top-down view, the tablecloth landmarks shift into the orientation that matches the fourth original photo. That means the fourth photo is the normal-perspective view looking in the direction that is to the right of the bottle in the third photo. In that aligned original image, the scene in front of the b... | Not supported with pagination yet | ||||
aa609e8da9a3fda93593b9daafa471eddde831705985606816a9c5ef103aaf92 | spatial/movement_direction_1k | train | Based on these two views showing the same scene: in which direction did I move from the first view to the second view? A. Diagonally forward and left B. Diagonally forward and right C. Directly right D. Directly left | null | A | I need to determine how the camera moved between the two viewpoints. In the two input photos, the key anchor objects are the pink toy train and the white rectangular case on the patterned cloth. Because perspective changes can be confusing from eye-level views, I will generate a bird’s-eye (top-down) view of the first ... | In the generated bird’s-eye view of the first scene, the toy train sits near the center of the dark square cloth, and the white case is positioned toward the upper-right area relative to the train. This top-down layout gives a clear reference for where each anchor lies. Next, I will generate a bird’s-eye view of the se... | In the generated bird’s-eye view of the second scene, the toy train remains near the center of the dark cloth, but the white case is now toward the upper-left relative to the train. That shift (upper-right to upper-left around the train) indicates a counter-clockwise change in viewpoint around the setup. Using the hint... | Not supported with pagination yet | ||||
a1fe64dd6c7f8e7a320f2dcb4c294a7a1deeae81409cc96960a4acddba3287d1 | spatial/movement_direction_1k | train | Based on these two views showing the same scene: in which direction did I move from the first view to the second view? A. Diagonally forward and left B. Directly left C. Directly right D. Diagonally forward and right | null | A | I need to infer how the camera shifted between the two viewpoints. In the first view, the metal bottle is near the center, with a black remote toward the left side and a white remote/AC controller near the lower-left edge; in the second view, the bottle is still central but the white remote appears on the right and mor... | This generated bird’s-eye view of the first scene shows the bottle near the middle of the cloth, the black remote positioned up-left of the bottle, and the white remote down-left of the bottle. With this top-down arrangement established, I will generate the bird’s-eye view for the second scene so I can compare how thes... | In the second bird’s-eye view, the bottle remains near the center, but the white remote has moved to the right side of the bottle, and the black remote is now down-left relative to the bottle. Comparing to the first top-down layout (white remote down-left; black remote up-left), the set of anchors around the bottle has... | Not supported with pagination yet | ||||
dae5e96b1b9769b8d0b1a7db49617f4d2e2568174580acda90f38598a8be5556 | spatial/movement_direction_1k | train | Based on these two views showing the same scene: in which direction did I move from the first view to the second view? A. Directly right B. Diagonally forward and right C. Directly left D. Diagonally forward and left | null | D | I need to infer how the camera translated between the two ground-level views. In both views, the key anchors are the long rectangular planter and the round stone pot beside it, plus the yellow curb/line on the ground. From the first view, the rectangular planter is seen broadside with the round pot on its left; in the ... | This generated bird’s-eye view (top-down) of the first viewpoint shows the rectangular planter lying horizontally across the scene, with a round pot at the left end and another round pot at the right end, aligned along a yellow line/curb. This clarifies that the camera in the first view is positioned on the near side o... | In the second bird’s-eye view, the same planter assembly is now oriented vertically in frame, with the yellow line/curb running along the right side of the planters, and the nearer round pot appearing at the bottom (closest to the camera). Relative to the first top-down view, the camera has shifted so that it is closer... | Not supported with pagination yet | ||||
b68b6e6359de705989ac65d2773bedf2da2a91999bed3dab99bf536cd20afbcb | spatial/movement_direction_1k | train | Based on these two views showing the same scene: in which direction did I move from the first view to the second view? A. Diagonally forward and left B. Directly right C. Diagonally forward and right D. Directly left | null | C | I need to infer how the camera moved between the two viewpoints by tracking stable anchor objects on the bed (yellow ball, purple cloth, white patterned cloth, red/white box, and the white/blue device). In the two perspective photos, these items shift positions relative to the frame, but perspective distortion makes it... | In the generated bird’s-eye view of the first viewpoint, the anchor layout is clear: the purple cloth sits toward the upper-left, the white/blue device is toward the upper-right, the red/white box is toward the lower-right, and the yellow ball is near the center. This gives a stable coordinate map of the scene. To comp... | In the generated bird’s-eye view of the second viewpoint, the same anchors appear rotated/shifted: the purple cloth is now on the right side (instead of upper-left), the white/blue device is now at the lower-right (instead of upper-right), and the red/white box is now at the lower-left (instead of lower-right), while t... | Not supported with pagination yet | ||||
36308001d2067f168898fe367b243638612d3bb56e6ee2792879a6ff9e3176f7 | spatial/movement_direction_1k | val | Based on these two views showing the same scene: in which direction did I move from the first view to the second view? A. Diagonally forward and right B. Diagonally forward and left C. Directly right D. Directly left | null | B | I need to infer how the camera moved between the two viewpoints. In the first view, the green ball on the patterned cloth is the main anchor, and the background changes from a sofa scene to a cabinet scene, suggesting the camera shifted position around the table rather than the table moving. To reason about direction (... | In the generated bird’s-eye view of the first viewpoint, the green ball sits near the center of a light floral square cloth. Around it, a prominent pink polka-dot patch lies on the left side of the circular cloth, and a large pink character-print region is on the right. This top-down layout gives me a stable reference ... | In the generated bird’s-eye view of the second viewpoint, the same anchor elements are present but their positions around the ball have rotated: the pink polka-dot patch that was on the left is now down toward the bottom, and the large pink character-print region that was on the right has shifted up/left relative to th... | Not supported with pagination yet | ||||
02844ad4bb56dc46468f2e194ce2613c9f119933c0548d41f5d3a56c86e01645 | spatial/positioned_direction_423 | train | Based on these four images (image 1, 2, 3, and 4) showing the blue ball from different viewpoints (front, left, back, and right), with each camera aligned with room walls and partially capturing the surroundings: If I was positioned where the light-colored sofa is and facing the same direction, what would be to the rig... | null | B | The question asks me to imagine standing where the light-colored sofa is, while keeping the same facing direction as that sofa view. So I first compare the room cues around the blue ball across the four views: the light-colored sofa appears in one direction, the dark brown sofa in another, the white TV cabinet with a s... | In the first generated top-down view, the ball stays at the center while the patterned cloth around it is seen from directly above, which helps fix the imagined viewpoint without perspective distortion. This corresponds to standing at the light-colored sofa side and looking toward the ball. But the question is not aski... | The second generated top-down view matches the original right-side room view, the one showing the white-red cabinet beyond the ball. Using that matched original image to identify the furniture clearly, the object to the right of the blue ball from the light-colored sofa’s viewpoint is the white-red cabinet. Therefore t... | Not supported with pagination yet | ||||
8566f6445fbf0ec1b62997075201df59ca1d33e842b111e7263930a443df1540 | spatial/positioned_direction_423 | val | Based on these four images (image 1, 2, 3, and 4) showing the black sneaker from different viewpoints (front, left, back, and right), with each camera aligned with room walls and partially capturing the surroundings: If I was positioned where the light purple sofa is and facing the same direction, what would be to the ... | null | B | The question asks me to imagine standing where the light purple sofa is and facing the same direction as the sofa-facing view. From the room clues across the sneaker’s front, side, and back views, the sofa is on one side of the bed while the TV, dining area, and window side occupy other directions around it. To answer ... | In the first generated top-down view, the sneaker is seen from above with its toe pointing to the right side of the frame, which matches the orientation when viewed from the sofa side. This establishes the imagined observer as being at the sofa, looking across the bed toward the sneaker in that same direction. Since th... | The second generated top-down view corresponds to the original right-side viewpoint of the sneaker. In that matching original view, the surroundings visible beyond the sneaker are the dining area, especially the wooden dining table. The TV appears in a different direction, and the brown curtains and windows are seen fr... | Not supported with pagination yet | ||||
8f610b4f1e6b35f4f99e30a2d17123ab0321b429cf655aaa96339577e2b1eb85 | spatial/positioned_direction_423 | val | Based on these four images (image 1, 2, 3, and 4) showing the blue stool and pillows from different viewpoints (front, left, back, and right), with each camera aligned with room walls and partially capturing the surroundings: If I was positioned where the two single sofas is and facing the same direction, what would be... | null | B | The question asks me to imagine standing where the two single sofas are, while keeping the same facing direction as they have in the room, and then determine what lies to the left of the blue stool and pillows from that imagined viewpoint. From the room views, the blue stool with folded pillows is near the center, and ... | In the first generated top-down view, the stool and pillows are centered, and the pair of single sofas are clearly on the far side of the room, facing toward the stool. This establishes the imagined observer as being at that side of the room looking inward toward the center. From that orientation, the question asks for... | The second generated top-down view matches the room arrangement seen in the left-side original view. In that aligned view, the object lying to the left of the blue stool and pillows from the imagined position at the two single sofas is the TV, which appears along that side of the room beside the media stand. So the cor... | Not supported with pagination yet | ||||
f6f8797b192786413355ab0d89fa4189991097efc32e915dfe68433ad74e6065 | spatial/positioned_direction_423 | val | Based on these four images (image 1, 2, 3, and 4) showing the black shoe from different viewpoints (front, left, back, and right), with each camera aligned with room walls and partially capturing the surroundings: If I was positioned where the light purple sofa is and facing the same direction, what would be to the lef... | null | A | The question asks me to imagine standing where the light purple sofa is and facing the same direction as the sofa-facing view. So I need to relocate the observer to the sofa side of the shoe, while keeping the viewing direction consistent with that position. From the four views, the sofa appears behind the shoe in the ... | In the first generated top-down view, the shoe is seen from above from the sofa-side position, which fixes the imagined observer at the sofa and facing toward the shoe. This view clarifies the shoe's toe direction relative to the shirt and nearby objects, but a top-down view from that position does not yet make the sho... | The second generated top-down view corresponds to the original right-side room view of the shoe. In that matching original view, the area to the left of the shoe is the bright window side with the brown curtains. The TV is seen in a different direction, and the wooden dining table appears in the opposite hallway-facing... | Not supported with pagination yet | ||||
0ab429b21185065a998ccd06acfb2259633ff40fb599505456dbdbcc260df885 | spatial/positioned_direction_423 | val | Based on these four images (image 1, 2, 3, and 4) showing the white slipper from different viewpoints (front, left, back, and right), with each camera aligned with room walls and partially capturing the surroundings: If I was positioned where the white slipper is and facing the same direction, what would be to my right... | null | C | The task is to imagine standing where the white slipper is, while keeping the slipper’s facing direction. Then I need to determine what would be on my right. From the four views, the slipper is shown from different sides along with room features: one side shows the window and wall, another shows the washing machine are... | In the first generated top-down view, the slipper’s opening is at the lower end and the toe points upward, so the imagined observer at the slipper is facing toward the top of that top-down layout. This establishes the observer’s orientation unambiguously. Since the question asks for what is to the right of that observe... | The second generated top-down view corresponds to the original view taken from the slipper’s right side. In that matching original image, the background on that side includes the washing machine. Cross-checking with the other original views helps separate it from the window-wall side and the curtain side. Therefore, th... | Not supported with pagination yet | ||||
81c14246a8d70744889adeecebe99b7400827f20a86f6591b0b1121a048eb8c2 | spatial/turn_degrees_204 | val | These four images (image 1, 2, 3, and 4) show the same scene from different viewpoints. Based on these four images: If I am standing at the same spot and facing the same direction as shown in image 4 and turn 180 degrees around, what is to my right? A. Blue cushions B. Door C. Yellow sofa D. Row of chairs | null | B | From the queried view, I can see the orange wall with the yellow sofa straight ahead, while the glass wall is off to the left edge. To answer what would be on the right after turning around, I first need a bird's-eye view centered on this viewpoint so the current facing direction is unambiguous before applying any turn... | In the first generated bird's-eye view, the yellow sofa sits along the far side of the room relative to the current viewpoint, matching the fact that image 4 faces toward the orange wall and sofa. The glass wall and yellow-framed door are on the left side of this top-down layout, and the blue cushions are on the opposi... | The turned top-down view now faces toward the blue cushions and glass wall side of the room, exactly opposite the original sofa-facing direction. From this turned viewpoint, the question asks for what is to my right. Using the ego-aligned top-down convention, a right query corresponds to an anticlockwise 90-degree rota... | The final aligned top-down view points toward the glass wall with the yellow-framed door. That orientation matches the original view showing the door and glass partition directly ahead. In that matching original view, the prominent object in front is the door. Therefore, after standing as in image 4, turning 180 degree... | ||||
f1475c6c69768986cdfbcf7659305cee281f22d01b9cc6ce875589e37c6c060d | spatial/turn_degrees_204 | val | These four images (image 1, 2, 3, and 4) show the same scene from different viewpoints. Based on these four images: If I am standing at the same spot and facing the same direction as shown in image 4 and turn 90 degrees to the right, what is to my behind? A. Several bookcases B. Bookcase C. Long table D. Card catalog c... | null | C | From the queried view in image 4, I can already note three key landmarks: the card catalog cabinet is on the right, a bookcase is on the left, and the corner of the long table is at the lower left. To answer a turn-and-direction question reliably, I should first form a bird's-eye view of this queried viewpoint so the c... | In the first generated bird's-eye view, the long table runs through the middle of the room with bookcases lining both sides, matching the study area seen across the input images. This overhead layout makes the queried orientation easier to track. The question says to turn 90 degrees to the right, so in this ego-aligned... | The second generated top-down view shows the room after that right turn: the table is now oriented across the frame, with the surrounding bookcases and open floor repositioned consistently around the observer. Now I apply the asked final self-relative direction, which is 'behind.' In this top-down convention, 'behind' ... | The final aligned top-down view matches the orientation of the original view that looks straight along the long table. In that matching original image, the object directly in front is the long table, with chairs along it and bookcases on both sides. Since the front of this final aligned view corresponds to what would b... | ||||
93950cb7622ac278094a8c1056b241b33875d6844fc21d7abbb534ff8fe708b0 | spatial/turn_degrees_204 | val | These four images (image 1, 2, 3, and 4) show the same scene from different viewpoints. Based on these four images: If I am standing at the same spot and facing the same direction as shown in image 1 and turn 180 degrees around, what is to my left? A. Pathway B. Bookcase C. A row of plastic boxes D. Wooden desk | null | C | From the queried view, I can already identify the main landmarks: tables and chairs ahead, a bookcase along the left edge, and plastic storage boxes along the right edge beside the railing. To answer a turn-and-direction question reliably, I should first form a bird's-eye view of this starting viewpoint so the current ... | In the first generated bird's-eye view, the aisle runs straight ahead, with the bookcase on one side and the row of plastic boxes on the opposite side. This confirms the starting orientation seen in the queried view. The question says to turn 180 degrees around, so I now apply a 180-degree rotation in the top-down fram... | The turned bird's-eye view now represents standing in the same spot but facing the opposite way. From this turned viewpoint, the question asks what is to my left. Using the stated top-down convention, a left query is handled by a clockwise 90-degree rotation relative to the turned viewpoint. So I generate one more alig... | The final aligned bird's-eye view matches the orientation of the view that looks directly toward the line of storage bins along the railing. In that matching original view, the object straight ahead is clearly the row of plastic boxes. Because this final alignment converts the asked-left direction into front, the thing... | ||||
4a30e536256954dcfda66c5723590840ffbabc33fa6de57784d96267c2fc564a | spatial/turn_degrees_204 | val | These four images (image 1, 2, 3, and 4) show the same scene from different viewpoints. Based on these four images: If I am standing at the same spot and facing the same direction as shown in image 3 and turn 90 degrees to the left, what is to my left? A. Black cabinet B. Podium C. Blackboard D. Rows of desks | null | B | From the queried classroom view, I can see the tall black cabinet beside the curtained windows, while the other views show the blackboard, podium, and rows of desks from other sides of the same front area. To answer a turn-and-direction question reliably, I should first form a bird's-eye view centered on the viewpoint ... | In the first generated bird's-eye view, the observer stands near the front teaching area with the blackboard along one side, the podium and chairs nearby, the desk rows off to the opposite side, and the black cabinet near the windows. This top-down layout makes the starting orientation from the cabinet-facing scene exp... | The turned bird's-eye view now represents the orientation after that left turn. From this turned viewpoint, the question asks what is to my left. To align that requested left-side direction with the forward direction for easier reading, I apply the same convention again: a left query corresponds to a clockwise 90-degre... | The final aligned top-down view matches the orientation of the view that looks toward the teaching station and blackboard area from the side. In that matching original scene, the object lying ahead in the aligned direction is the podium rather than the cabinet, blackboard, or desk rows. Therefore, after starting as in ... | ||||
65373484c0c424330b93d10e8defe10b3c722b8db9a178851245292811642a42 | spatial/turn_degrees_204 | val | These four images (image 1, 2, 3, and 4) show the same scene from different viewpoints. Based on these four images: If I am standing at the same spot and facing the same direction as shown in image 2 and turn 90 degrees to the left, what is to my left? A. Row of chairs B. Blue cushions C. Yellow sofa D. Door | null | C | From the queried view, I can see the blue cushions ahead, chairs along the left wall, and the door on the right, while the other views show the rest of the room including the yellow sofa. To keep the current standing position and facing direction straight before making any turn, I will first form a bird's-eye view of t... | In the first generated bird's-eye view, the same arrangement becomes clear from above: the blue cushions are in front of the current viewpoint, the row of chairs runs along the left side, the glass wall and door are on the right, and the yellow sofa is behind. The question says to turn 90 degrees to the left, so in thi... | The turned top-down view now represents standing in the same place after that left turn. From this new orientation, the question asks what is to my left. Using the same top-down convention, a left query means rotating clockwise by another 90 degrees relative to the turned viewpoint. I generate one more aligned bird's-e... | The final aligned bird's-eye view has the same orientation as the view that looks toward the orange wall with the yellow sofa. Checking the corresponding original image confirms that the prominent object straight ahead in that direction is the yellow sofa. Therefore, after starting as in image 2, turning 90 degrees lef... | ||||
9e9b3ae7f7a309a92479fb4e8014d84950d8aeb67651b5e57713d8dcda056a65 | spatial/turn_move_closer_2k | train | Based on these four images (image 1, 2, 3, and 4) showing the jam bottle from different viewpoints, with each camera aligned with room walls and partially capturing the surroundings: If I am standing at the same spot and facing the same direction as shown in image 2, then I turn right and move forward, will I get close... | null | B | The question starts from the viewpoint shown in image 2, where the jam bottle is in front of a plain white wall. The target object is the white cabinet seen in another view of the same table. To answer a turn-and-move question, I first need a bird's-eye view of the starting viewpoint so I can fix the observer's current... | In the first generated bird's-eye view, the tabletop pattern around the bottle makes the starting orientation from image 2 clear. Since the question says to turn right, I apply an anticlockwise top-down rotation to the observer's facing direction. That tells me which side of the table and room will be in front after th... | The turned bird's-eye view shows the orientation after that right turn, and this new facing direction aligns with the scene where the white cabinet is ahead of the bottle. That matches the room view with the cabinet and lounge area, confirming that moving forward from the image-2 position after turning right would head... | Not supported with pagination yet | ||||
5e2a7b52ee5ddb8ad9edae8efcbfbbb45922edb8617698e0fde4d942b25810d5 | spatial/turn_move_closer_2k | train | Based on these four images (image 1, 2, 3, and 4) showing the white bottle from different viewpoints, with each camera aligned with room walls and partially capturing the surroundings: If I am standing at the same spot and facing the same direction as shown in image 1, then I turn right and move forward, will I get clo... | null | A | The question uses the viewpoint from image 1 as the starting position and asks about the pink plush toy and the headboard. In that starting photo, I can see the bottle on the bed, a window and curtain ahead, and the plush toy/headboard are not directly in front from this view, so I need to establish the observer’s curr... | In the first generated bird’s-eye view, the bottle sits near the middle of the red cloth, with the surrounding items laid out around it. This top-down perspective clarifies the orientation of the starting view from image 1. Since the question says to turn right, I now update the facing direction accordingly; in this se... | The turned bird’s-eye view shows the new forward direction after the right turn. From that orientation, moving forward points toward the side of the bed where the pink plush toy and the padded headboard appear in the room views. This matches the direction seen in the original view that includes the plush toy and headbo... | Not supported with pagination yet | ||||
0df4a80b4a31832e8334ee38a517a3b82def00f4c2709471478f0ebeae84b62f | spatial/turn_move_closer_2k | train | Based on these four images (image 1, 2, 3, and 4) showing the yellow ball from different viewpoints, with each camera aligned with room walls and partially capturing the surroundings: If I am standing at the same spot and facing the same direction as shown in image 1, then I turn left and move forward, will I get close... | null | B | The question starts from the viewpoint in image 1, and the target object to compare distance to is the wall. In that view, the camera is facing the wall with the dark hood, while the yellow ball sits on the bed in front. To reason about what happens after turning left and then moving forward, I should first form a bird... | The first generated bird's-eye view shows the layout around the yellow ball from the starting orientation: the cigarette pack is toward the upper-left, the green toy vehicle toward the upper-right, the white cloth to the right, and the white-blue object near the lower edge. This top-down arrangement helps anchor which ... | In the turned bird's-eye view after the left turn, the scene has rotated so that the new forward direction aligns with the side that was to the observer's left in image 1. Comparing the room views, that leftward-facing orientation matches the plain tiled wall seen in image 4. Since facing that direction means the obser... | Not supported with pagination yet | ||||
f2d9375caa7d9cf9e4fa7b6c5c3c8df3ab22bf9977a550063baa0e46f7ea2b2c | spatial/turn_move_closer_2k | train | Based on these four images (image 1, 2, 3, and 4) showing the black boot from different viewpoints, with each camera aligned with room walls and partially capturing the surroundings: If I am standing at the same spot and facing the same direction as shown in image 3, then I turn left and move forward, will I get closer... | null | B | The question starts from the viewpoint shown in image 3. The target to compare distance to is the wall and window. From that starting photo, I can already tell the boot is being viewed from one side, with the room corner and fixtures behind it, but to answer a turn-and-move question reliably I need to place that viewpo... | In the first generated bird's-eye view, the boot's toe points toward the left side of the table, while the nearby bottle and small items line up consistently with the starting scene. This top-down layout clarifies where the observer is standing relative to the boot in image 3. For this task, turning left from the start... | In the turned bird's-eye view, the observer is now facing the side of the setup that matches the window-side orientation, consistent with the original view where the wall and window are behind the boot. After the left turn, moving forward would head toward that side of the room rather than away from it. So the forward ... | Not supported with pagination yet | ||||
7c9e912f900dee199a7ad2a6608b831939713c82ad6d4a6bd9d85916404249ce | spatial/turn_move_closer_2k | train | Based on these four images (image 1, 2, 3, and 4) showing the white football from different viewpoints, with each camera aligned with room walls and partially capturing the surroundings: If I am standing at the same spot and facing the same direction as shown in image 2, then I turn left and move forward, will I get cl... | null | A | The question starts from the viewpoint shown in image 2, and the target object is the window. From that view, I can see the ball on the green cloth, with the sofa and bright wall area behind it, but I need to pin down the observer’s current facing direction relative to the whole setup before applying the left turn. So ... | In the first generated bird’s-eye view, the layered cloths around the ball become clear: the blue striped piece lies beyond the green cloth on one side, the white floral cloth is on the adjacent side, and the red striped cloth extends on another side. This top-down arrangement lets me anchor the starting viewpoint from... | In the turned bird’s-eye view, the observer’s new forward direction lines up with the side of the setup that corresponds to the window-facing scene. Comparing this with the original photos, that turned orientation matches the view where the window is in front of the ball. So after turning left, moving forward would tak... | Not supported with pagination yet | ||||
6592c0a9e50ee79e4069f5674ca73f21ba95a4043a6e9e8df84575829a9f978b | spatial/multiview_58k | train | Treat the projected screen as being west of the red wall lights. In which direction are the floor candles relative to the red wall lights? | A: West, B: South, C: East, D: North | A | The two camera views show overlapping parts of the same dark room. One view includes the projected screen and the red wall lights together, while the other shows the floor candles and the red wall lights more clearly. To answer the direction question, I need the full left-to-right layout of these landmarks in one consi... | In the merged panoramic view, the projected screen appears on the left side, the cluster of floor candles sits near the middle, and the red wall lights are on the right wall. The question tells us to treat the projected screen as west of the red wall lights, so left corresponds to west and right corresponds to east. Si... | Not supported with pagination yet | Not supported with pagination yet | ||||
7189891838ef5d6f9eea0ed5cd8a3eec2994aed9cf4a9cd283799d79d0594c72 | spatial/multiview_58k | train | The pictures are taken continuously from a first-person perspective. In which direction is the camera rotating? | A: Right, B: Backward, C: Left, D: Forward | C | The two first-person views show the camera looking through a doorway into a bright room with large windows. In one view, the open door and the left portion of the window wall are more prominent; in the other, the right doorframe and a slightly more centered view of the window wall are visible. To determine rotation dir... | The merged panoramic view shows a continuous layout: the open door and left side of the room appear on the left, then the central wall/doorframe, and then the brighter right side of the room with the window bench and objects near the far-right window. This means the second view is shifted leftward relative to the first... | Not supported with pagination yet | Not supported with pagination yet | ||||
76515db376d1402154450e16f345ae62d893af1464e47f65fed2cc056d04cc27 | spatial/multiview_58k | train | When you took the second photo, where was the bathroom area with the shower and sink relative to you? | A: Rear right, B: Front right, C: Front left, D: Rear left | B | The two photos show adjacent parts of a compact room suite. One view looks into the sleeping area with the bed, while the other looks into the bathroom area with the shower enclosure and round sink. To answer where the bathroom was relative to the camera when the second photo was taken, I need to combine these partial ... | In the merged panoramic view, the bedroom occupies the left/center portion, and the bathroom area with the sink and glass shower is clearly on the right side of the doorway zone. The second photo is the bedroom-facing shot, so the camera for that photo is positioned near the bathroom/entry side looking forward into the... | Not supported with pagination yet | Not supported with pagination yet | ||||
6afe4c30729327ead7818d05ba5be32ff8e6496672c66d0972dc7e6adaa53eaa | spatial/multiview_58k | train | The pictures are taken continuously from a first-person perspective. In which direction is the camera rotating? | A: Forward, B: Right, C: Left, D: Backward | B | The first-person views show the same interior from two nearby angles. In the earlier view, most of the frame is a bright white wall with window-light rectangles and a tan sliding panel near the center-right. In the later view, the scene opens into the hallway and seating area, with the shelving, corridor, painting, and... | The merged panoramic view reveals a continuous scene: the bright white wall is on the far left, then the tan panel/door frame, and farther right are the shelves, open doorway, hallway, painting, and orange chair. This means the first input image is looking more toward the left side of the room, while the second input i... | Not supported with pagination yet | Not supported with pagination yet | ||||
f5b6575d24ed4f69d86a3077b577c6f334982aca38b7e759a19957d864329df1 | spatial/multiview_58k | train | If the direction toward the window is considered the front, where is the hallway opening relative to the large green wall panel? | A: Back, B: Right, C: Front, D: Left | D | The scene shows a large green stone-like wall panel spanning most of the wall, with a window visible in the second view and an opening toward a hallway or adjacent passage visible on the left side of the first view. Because the two input views split the room layout, I will generate a merged panoramic view next so I can... | In the merged panoramic view, the large green wall panel is centered, the window is clearly on the right side, and the hallway opening is on the left side of the panel. If the direction toward the window is treated as the front, then facing front means facing rightward in this panorama. When facing that way, the left s... | Not supported with pagination yet | Not supported with pagination yet | ||||
8acb9b8d72e464015ea3a2b9acb7e7fbf4a362245d9760f3dd79af48dce603ef | spatial/depth_10k | train | Is grey knit dress at center behind matte black jacket at left? | (A) yes (B) no | B | I need to determine the front-back relationship between the grey knit dress in the center area and the matte black jacket on the left. From the original photo, both adults overlap slightly, so I will generate a depth-support view next to compare which person is closer to the camera based on their relative depth shading... | The generated depth-support view shows the left person in the matte black jacket mostly in cooler teal tones, while the grey knit dress wearer appears in relatively warmer green tones over the torso and lower body. In this depth visualization, the warmer tones indicate being closer to the camera than the cooler tones. ... | Not supported with pagination yet | Not supported with pagination yet | ||||
c3ee9a8a7773378ac510c660c9ee89d57b5028cdd78b27aa729e26346f07f45c | spatial/depth_10k | train | Is white satin bed at center further to camera compared to white matte chest of drawers at upper left? | null | No, white satin bed at center is in front of white matte chest of drawers at upper left. | The question asks for a depth comparison between the white satin bed in the center-right foreground and the white matte chest of drawers at the upper left. From the room photo, the bed occupies much of the right side and appears to extend toward the camera, while the drawers sit farther back near the wall. To verify th... | In the depth-support view, the bed region on the right is shown in brighter yellow-green tones, while the chest of drawers area at the upper left is darker blue-purple. This contrast indicates the bed is closer to the camera and the drawers are farther away. Combining that with the original image layout, the white sati... | Not supported with pagination yet | Not supported with pagination yet | ||||
01149cc0de1ad71c877a0cce1fcc614cf7354bdd198e1a29fb1a0e7809c48325 | spatial/depth_10k | train | Estimate the real-world distances between objects in this image. Which object is farther from wine bottle (highlighted by a red box), bottle (highlighted by a blue box) or desk (highlighted by a green box)? | (A) bottle
(B) desk | A | The question asks which of the two marked objects is farther from the red-boxed wine bottle: the blue-boxed bottle or the green-boxed desk. From the original room view, all three are near the cabinet area, but perspective alone is not enough to judge their real-world depth separation confidently. I will generate a dept... | In the generated depth-support view, the cabinet region containing the red-boxed wine bottle and blue-boxed bottle falls in a similar darker blue/cyan depth band, while the surrounding nearer foreground elements shift toward greener/yellower tones. The desk highlighted in green corresponds to the cabinet/front furnitur... | Not supported with pagination yet | Not supported with pagination yet | ||||
87745178c590415f8b9caf5b8ab10e86ea90abf564f0866bf7e45b15034a28e3 | spatial/depth_10k | train | Can you identify the 2nd appliance behind the matte glass bottles at upper center, starting your count from the anchor? Your answer should be formatted as a tuple, i.e. [(x, y)], where the tuple contains the x and y coordinates of a point satisfying the conditions above. | null | [(0.622, 0.336)] | The question asks for the 2nd appliance behind the matte glass bottles at the upper center, counting from the anchor object. In the cabinet photo, the matte glass bottles are the small frosted-looking bottles on the top shelf near the middle-right. To resolve which appliance is behind them in depth order, I need a dept... | In the generated depth-support view, the upper-center shelf region shows the matte glass bottles as a cluster, and the appliance behind them appears slightly farther back in the same area. Matching this depth region back to the original image, the appliance is the beige square fan-like unit mounted behind the bottles o... | Not supported with pagination yet | Not supported with pagination yet | ||||
5ac42e2ac550ad7c02e91564ec2e006aed110613c937062fe99c810b9bf96456 | spatial/depth_10k | train | Estimate the real-world distances between objects in this image. Which object is closer to the stone clock tower at center? | (A) metallic clock face at upper center (B) brown stone clock tower at center | B | The question asks which object is closer to the stone clock tower at the center. From the original photo, one option is the metallic clock face mounted on that tower, and the other option is the brown stone clock tower itself. To resolve the spatial relation cleanly, I will generate a depth-support view of the scene so... | The generated depth-support view shows the tower and attached church building in a similar blue tone, while the foreground vegetation is greener/yellower and therefore closer, and the sky/background are darker purple and farther away. The metallic clock face lies directly on the front surface of the central tower and s... | Not supported with pagination yet | Not supported with pagination yet | ||||
88cd5be2f7707405f9dbef843905715887c371a127ad3598642d97d2b17bbc25 | spatial/counting_5k | train | What is the total number of buses visible in this image? | null | 5 | The street scene contains several large vehicles, and some buses are partially hidden behind the sightseeing bus and by the left edge of the frame. To avoid missing small or occluded bus portions, I will generate a mask overlay that isolates only the bus instances against a plain background, making each separate bus si... | The mask overlay cleanly separates the buses from the rest of the scene. I can identify one small partial bus at the far left edge, a second red double-decker just to its right, the large sightseeing bus in the center, a fourth bus segment visible behind the right side of the sightseeing bus, and a fifth bus at the far... | Not supported with pagination yet | Not supported with pagination yet | ||||
104da3220d2e2f26963faf285a9770ead5985dccda6ec830b779c58f0d0b81a9 | spatial/counting_5k | train | What is the number of sweaters visible in this image? | null | 2 | The scene is crowded, so some clothing items overlap and different tops could be confused with sweaters. I first identify likely sweater regions in the original image: the large brown argyle knit in the foreground is one clear candidate, and there may be another partial knit garment at the right edge. To verify the cou... | The mask overlay isolates the sweater regions and outlines each separate instance clearly. In this view, there is one large sweater occupying most of the center foreground and one additional smaller sweater fragment on the right side. Since these are distinct isolated instances, the total number of visible sweaters is ... | Not supported with pagination yet | Not supported with pagination yet | ||||
94bde2f3f5d462aef41c8ce04a461b5ab4b764e7fef921832b8a27032667be64 | spatial/counting_5k | train | What is the number of train_(railroad_vehicle) visible in this image? | null | 1 | In the seaside railway scene, the target category is train. The main challenge is making sure I count only actual train instances and not confuse the rails, platform, or background structures with separate vehicles. To verify the count cleanly, I will generate a mask overlay that isolates only the train instances again... | The generated mask overlay isolates the train from the rest of the scene. In this extracted view, there is one continuous train instance visible and no second separate railroad vehicle elsewhere in the image. Combining this with the original scene confirms the count is 1. | Not supported with pagination yet | Not supported with pagination yet | ||||
b9dcad8070982099b005258109f25454b3284f14e60d4ceda56910991356bccf | spatial/counting_5k | train | What is the number of dogs visible in this image? | null | 1 | The scene is a close-up of a small dog wearing a pink life jacket. Because the framing is tight and parts of the body are cropped, I want to generate a mask overlay that isolates only the dog from the surrounding people, clothing, and accessories. That will make it easier to verify whether there is just one visible dog... | The generated mask overlay isolates a single continuous dog instance: one head with two ears, one muzzle, and one visible section of the body/neck below it. No second separated dog-shaped region appears anywhere else in the overlay. Combining this with the original photo, the number of dogs visible is 1. | Not supported with pagination yet | Not supported with pagination yet | ||||
3fdbe24e15c5defa455d8650c9d93a97dd7bad288cccb41f1331a86b3fb2ce2f | spatial/counting_5k | train | How many dogs can be seen in this image? | null | 1 | In the bedroom scene, I need to count dogs specifically. There is one clearly visible light-colored dog lying across the blankets, while another curled animal in the upper-left could be a different pet and might cause confusion if I rely only on the full scene. To avoid mixing the dog with non-target animals or bedding... | The mask overlay isolates the target category and shows a single distinct dog silhouette with black contouring around its body and head. No second dog-shaped instance appears in the overlay. Combining this with the original scene, the total number of dogs visible is 1. | Not supported with pagination yet | Not supported with pagination yet | ||||
11679a8af146960e1abe773598c3a567e719d2a3c931e08a832da8910b6379a0 | spatial/rotation_10k | train | Given the initial photo of an object, generate the views after performing the following camera rotations: Step 1: rotate 33° to the right; Step 2: rotate 80° to the left. | null | Step 1: rotate 33° to the right; Step 2: rotate 80° to the left | After rotating 33° to the right, the object now looks like this: | After rotating 80° to the left, the object now looks like this: | Not supported with pagination yet | |||||
a651ee0934f248fc2d5b5c0c8a9b962fc3db9d5809b4c063f10ec58327f595f9 | spatial/rotation_10k | train | Given the initial photo of an object, generate the views after performing the following camera rotations: Step 1: rotate 69° to the left. | null | Step 1: rotate 69° to the left | After rotating 69° to the left, the object now looks like this: | Not supported with pagination yet | Not supported with pagination yet | |||||
18fe7d77ec0473274ca22f0747e22cbef68bf4785a70ad04805c1e901e76ce7b | spatial/rotation_10k | train | Given the initial photo of an object, generate the views after performing the following camera rotations: Step 1: rotate 87° to the left. | null | Step 1: rotate 87° to the left | After rotating 87° to the left, the object now looks like this: | Not supported with pagination yet | Not supported with pagination yet | |||||
91f24373007f1c5c15c93e6abdfa775e27d25e40665e271fbc7becc2a48468d7 | spatial/rotation_10k | train | Given the initial photo of an object, generate the views after performing the following camera rotations: Step 1: rotate 51° to the right; Step 2: rotate 45° to the left. | null | Step 1: rotate 51° to the right; Step 2: rotate 45° to the left | After rotating 51° to the right, the object now looks like this: | After rotating 45° to the left, the object now looks like this: | Not supported with pagination yet | |||||
c153247f64851fea39815e584bb967b26bc5549d3e8683e01581e38271b48d00 | spatial/rotation_10k | train | Given the initial photo of an object, generate the views after performing the following camera rotations: Step 1: rotate 47° to the left; Step 2: rotate 83° to the left. | null | Step 1: rotate 47° to the left; Step 2: rotate 83° to the left | After rotating 47° to the left, the object now looks like this: | After rotating 83° to the left, the object now looks like this: | Not supported with pagination yet | |||||
2a60f0b468e0194412e79c41a43784ab5bfd874f34f129807bfbd6251bf8ecc4 | spatial/perspective_6k | train | Standing at the spot labeled "X" in the image, imagine you rotate 90° to your left. Relative to your new facing direction, where would the Toaster (near marker 1) be located?
Options: A: approximately straight in front of you, B: on your right | null | A | From the room view, the standing point “X” is on the floor near the inner corner of the L-shaped counter. The toaster is identified as the object near marker 1, which sits on the left run of the counter. To determine where the toaster ends up relative to the new facing direction after a 90° left rotation, I need to gen... | In the generated egocentric view from X after turning left, the toaster is directly ahead on the countertop, occupying the central forward area of the scene rather than appearing off to the right side. This indicates that, relative to the new facing direction, the toaster is approximately straight in front of you. Ther... | Not supported with pagination yet | Not supported with pagination yet | ||||
de4896988bd02c2ef98638beea12f8231bfe1a5ff66d65cf2add693cb6e098df | spatial/perspective_6k | train | Go to the location marked with an 'X' in the image. After you rotate 90° to your left, does the chair near label 2 become closer or farther away? Options: A: Further, B: Closer | null | B | In the room view, the marked X is on the floor near the right side of the bed. The chair associated with label 2 is part of the small desk-and-chair setup on the left side of the room. From the original viewpoint, that chair is across the room and offset to the left relative to X. To judge whether it becomes closer or ... | In the generated egocentric view from X after turning left, the desk and its chair are now directly in front of the viewer and occupy a larger portion of the scene, indicating a shorter line-of-sight distance compared with the original view where they were farther across the room. This shows that after moving to X and ... | Not supported with pagination yet | Not supported with pagination yet | ||||
245a79fdfc69a281b425e72aea31bdd587d1cbff18cd8aa1e8b4911064ad9ebc | spatial/perspective_6k | train | Seated at the spot labeled 'X' in the image, and then turning to face 90° to the left, would the Cup (near label 6) fall on your left side or your right side? Options: A: left, B: right | null | B | In the room view, the seated position is marked by the red “X” on the floor. The cup is referenced as being near label 6, which is located on the table area to the left side of the room from this original camera angle. However, “left vs right” must be judged after the person at X turns 90° left, so I need to generate a... | In the generated egocentric view from X after turning 90° left, the table/desk area (where the cup near label 6 would be) appears on the right side of the viewer’s field of view. That means once seated at X and turned left, the cup would be on the viewer’s right side, so the correct option is B (right). | Not supported with pagination yet | Not supported with pagination yet | ||||
2a7ca06d2fab97f136c8b7aa76346e76f091a4e9fba273e694c1095c4c285877 | spatial/perspective_6k | train | Sitting at the spot labeled 'X' in the picture, and then turning to face 90° to the left, would the doorway by marker 15 end up on your left side or your right side?
Options: A: right, B: left | null | A | From the room view, I locate the sitting point marked ‘X’ on the floor near the lower-right. The doorway of interest is the open double-door on the far-left side of the scene, labeled with marker 15. To determine whether that doorway ends up on the left or right after turning 90° left from the orientation implied at X,... | In the generated egocentric view from X after turning 90° left, the open double-doorway (the one corresponding to marker 15 in the original view) appears on the right side of the field of view. Since objects on the right side of this egocentric view correspond to being on your right side after the turn, the doorway by ... | Not supported with pagination yet | Not supported with pagination yet | ||||
d82a7f86722c636b938e9b722b39b3bd59a7bfe276965f1c96884ab9b72dba2b | spatial/perspective_6k | train | Standing on the point labeled 'X' in the image, if I rotate 90° to the right, does the ShelvingUnit (near label 16) move nearer to me or farther away? Options: A: Closer, B: Further | null | A | From the room view, the red X is on the floor near the bottom-center. The ShelvingUnit labeled 16 is on the right side of the scene, tucked against the wall near labels 14 and 17. To judge whether it becomes nearer or farther after a 90° right rotation (a change in facing direction but not position), I need to generate... | In the generated egocentric view from X after turning 90° right, the shelving unit appears at the left edge of my view, very close to my position (it’s now alongside me rather than across the room). Compared with the original view where it sat off to the right and more across the space, this indicates it is nearer to m... | Not supported with pagination yet | Not supported with pagination yet | ||||
d9387657831489ac3c334215473300d049f159f85ebf99c8b9e1573a8f061cc2 | spatial/directional_1k | train | Consider the real-world 3D locations and orientations of the objects. If I stand at the couch's position facing where it is facing, where is the black bookcase? | A: left, B: front, C: right, D: back-left | B | In the original room view, I can identify the couch on the right side and the black bookcase along the left wall, but raw RGB alone can make exact instance boundaries and the anchor-target pairing a bit ambiguous in a cluttered indoor scene. To localize the relevant objects precisely before judging their 3D relation, I... | In the masked third-person view, I add a red mask to the black bookcase and a yellow mask to the couch so the two key objects are isolated clearly. The red-masked bookcase occupies the left side of the room near the wall, while the blue-masked couch sits on the right side facing inward toward the open floor area. This ... | The directional support view adds an orientation cue at the couch, which clarifies the couch-centered forward axis. The cue shows that if I stand at the couch and face the same way it faces, the forward direction projects out into the room toward the area where the bookcase stands, rather than toward the camera-side fu... | In the egocentric rendered view from the couch's position, the black bookcase appears straight ahead in the field of view rather than off to either side. That means relative to the couch's facing direction, the black bookcase is in front. | ||||
1780693c9585fcdb6b66bb26c4550284b2ebc5c6fa2ce81f031e33221e3def2e | spatial/directional_1k | train | Consider the real-world 3D locations and orientations of the objects. If I stand at the man's position facing where it is facing, where is the white baseball bat? | A: front, B: right, C: back-left, D: front-left | D | I first inspect the original baseball scene to identify the anchor person and the queried object, the white baseball bat. In the raw RGB view, the players, umpire, and bat overlap heavily, so the exact anchor location and the bat's extent are somewhat ambiguous. To localize the relevant instances precisely, I will gene... | In the masked third-person view, I add a red mask to the man, which cleanly marks the anchor person's body near the center of the scene. The white baseball bat is visible crossing in front of him toward the catcher on the right side of the camera view. This makes the anchor-target pair clear in the shared scene. With t... | The directional support view places orientation cues on the red-masked man, clarifying the anchor-centered axes. The forward axis extends away from the man's position, and the side axis shows which direction is to his left versus right when adopting his stance. Combining that orientation with the bat's placement in the... | In the egocentric rendered view, the baseball bat appears in the forward field and shifted to the left side relative to the man's facing direction. That confirms the bat is at the man's front-left. Therefore the correct relation is front-left. | ||||
e5b86424e32842aa5f5ab8e22823d387ae80e8c720827806077571ec15f3bc19 | spatial/directional_1k | train | Consider the real-world 3D locations and orientations of the objects. If I stand at the child's position facing where it is facing, where is the yellow toy? | A: right, B: back, C: left, D: front | A | In the original dining-table scene, the child is the anchor and the yellow toy is the target, but raw RGB alone can make the exact instance boundaries and identities a little ambiguous because of clutter, overlap, and perspective. I will first generate a masked third-person view so I can localize the child and the toy ... | In the masked third-person view, I add a red mask to the yellow toy and a yellow mask to the child. This makes the relevant instances clear: the child is seated at the table near the center-right, and the toy is across from the child on the left side of the camera view. That left/right wording here is only an identity ... | The directional support view places an orientation cue at the child and shows the child's facing direction extending toward the right side of the scene. This clarifies what counts as straight ahead for the child, which is essential because the camera's viewpoint is different from the child's viewpoint. With the child-c... | In the egocentric rendered view from the child's position, the toy appears ahead and to the right of the child's forward direction rather than directly in front or off to the left. Combining that with the earlier localization and facing cue, the yellow toy is at the child's front-right. | ||||
f29c6d79540a8dfabad41f4bdfd245a5a0b13378f7455fd21eadb8b55603663f | spatial/directional_1k | train | Consider the real-world 3D locations and orientations of the objects. If I stand at the dog's position facing where it is facing, where is the yellow pot? | A: left, B: back, C: front-right, D: front | C | From the original scene, I can identify the dog near the doorway and a pot beside it on the deck, but raw RGB alone can make exact instance boundaries and overlap a bit ambiguous. To localize the anchor and target precisely before judging direction, I will generate a masked third-person view that highlights the dog and... | In the masked third-person view, I add a red mask to the yellow pot and a yellow mask to the dog. This makes the two relevant objects easy to isolate: the red-masked pot sits just to one side of the yellow-masked dog in the camera view, with the pot slightly higher in the scene and overlapping the dog's head area. That... | The directional support view places orientation axes on the dog, which clarifies how the dog is facing in 3D rather than just where it appears in the photo. With that forward reference, the pot is not merely beside the dog in the image; it lies ahead of the dog's facing direction and offset to the dog's right side. To ... | In the egocentric rendered view, the pot appears in front of the dog and shifted to the right side of the dog's forward line. Combining this with the earlier localization and orientation cue, the yellow pot is at the dog's front-right. | ||||
729d29ad8200ed9848aef73fd21b01600fa0c9ac97756844a598b1d0f3af97e8 | spatial/directional_1k | train | Consider the real-world 3D locations and orientations of the objects. If I stand at the woman's position facing where it is facing, where is the blue water? | A: front-left, B: back-left, C: right, D: front | A | In the original promenade scene, the woman near the railing is the anchor, and the blue water is the target. From the raw RGB view alone, object boundaries and the exact extent of the water are a little ambiguous because of distance, people, and the stone barrier, so I will generate a masked third-person view to locali... | In the masked third-person view, I add a red mask to the woman, which cleanly marks the anchor standing beside the stone railing. I also add a yellow mask to the water area on the right side of the scene beyond the railing, which identifies the queried target. The woman's placement and body orientation are now clearer:... | The directional support view adds an orientation cue at the woman's position, showing the direction she is facing and a sideways reference for her left-right axis. This clarifies that her forward direction runs away from the camera along the promenade, while the water lies off to one side beyond the railing. Combining ... | In the egocentric rendered view, the promenade structure appears ahead while the open side toward the water falls to the left of the woman's forward direction. Reading the target from this anchor-centered perspective places the blue water at the front-left. | ||||
14549237c82a83a555289c01848415799dd3002e9aa33db12e089885aafd9ef6 | spatial/undirectional_5k | train | Consider the real-world 3D locations of the objects. If I stand at the exhaust hood's position and look at a point roughly between the black microwave and black kitchen utensil, where does the black microwave appear to be? | A: bottom, B: left, C: right, D: top | D | From the kitchen photo, the relevant objects are the exhaust hood above the stove, the black microwave on the upper left, and the black kitchen utensil resting on the left countertop. Because the raw RGB view is cluttered and perspective can make exact object extents ambiguous, I will first generate a masked third-pers... | In the masked third-person view, I add a blue mask to the exhaust hood across the upper center, a red mask to the black microwave in the upper-left cabinet area, and a yellow mask to the black kitchen utensil on the left countertop. This confirms the anchor is the hood, the target is the microwave, and the utensil is t... | The directional support view from the exhaust hood shows the microwave and the utensil separated mainly along a vertical axis, with the microwave above and the utensil below. Looking toward the midpoint between them sets the forward reference between those two objects, and from that anchor-centered setup the microwave ... | Not supported with pagination yet | ||||
08a5b800321816773a24c63a29ed4620ddf474c8bfb18458cd641c255d5c3113 | spatial/undirectional_5k | train | Consider the real-world 3D locations of the objects. If I stand at the computer monitor's position and look at a point roughly between the white keyboard and yellow mouse, where does the white keyboard appear to be? | A: right, B: back, C: front, D: left | A | From the original desk photo, I can identify the computer monitor at the back center, the white keyboard near the front, and the yellow mouse on the right side of the desk. But raw RGB alone can make exact instance boundaries and the intended target point a little ambiguous in a cluttered workspace, so I will generate ... | In the masked third-person view, I add a red mask to the white keyboard, a yellow mask to the computer monitor, and a blue mask to the yellow mouse. This makes the key objects unambiguous: the yellow monitor is the anchor, the red keyboard is the queried object, and the blue mouse helps define the point roughly between... | In the directional support view from the monitor position, the keyboard and mouse are both out in front of the anchor, and the orientation cue clarifies the monitor-centered forward axis toward the desk surface. Relative to the point between the keyboard and mouse, the keyboard lies on the right-hand side in this ancho... | Not supported with pagination yet | ||||
e1cff971f54b7662b8d57a505a14050d9007bd454ea046fd07a292392bbd6193 | spatial/undirectional_5k | train | Consider the real-world 3D locations of the objects. If I stand at the window's position and look at a point roughly between the gray computer monitor and gray keyboard, where does the gray computer monitor appear to be? | A: right, B: left, C: front, D: back | B | From the original desk scene, I can identify the window on the left, the gray computer monitor on the desk ahead, and the gray keyboard in the foreground. But raw RGB alone can be a little ambiguous about the exact object extents and centers, especially with clutter and perspective. So I will generate a masked third-pe... | In the masked third-person view, I add a red mask to the window, a yellow mask to the gray keyboard, and a blue mask to the gray computer monitor. This makes the three key objects unambiguous: the red window is along the left wall, the yellow keyboard is on the near tabletop, and the blue monitor is farther ahead near ... | The directional support view from the window position shows the monitor and keyboard separated in depth, with the viewing direction aimed toward the midpoint between them. In this anchor-centered setup, the monitor falls to the left side of that forward line rather than the right. Therefore, the gray computer monitor a... | Not supported with pagination yet | ||||
28be4a91a225c7b9ed62be574003292d18fb857402a17f251f70ebb047e8f167 | spatial/undirectional_5k | train | Consider the real-world 3D locations of the objects. If I stand at the baseball's position and look at a point roughly between the black baseball bat and black catcher, where does the black baseball bat appear to be? | A: bottom, B: top, C: right, D: left | B | In the original baseball scene, the key objects are the baseball out on the field, the black baseball bat near the batter, and the black catcher behind home plate. Because small sports objects can be hard to localize precisely in raw RGB and the bat/catcher overlap with other players, I will generate a masked third-per... | In the masked third-person view, I add a red mask to the baseball and a blue mask to the black catcher, which makes the anchor and the reference pair easy to identify. The black baseball bat is the dark bat raised above the batter. This confirms the baseball is well out in front-left of home plate in the camera view, w... | In the directional support view from the baseball, the catcher appears near the center as the forward reference, and the black baseball bat sits above that central aiming point. That means when I stand at the baseball and look roughly between the bat and catcher, the bat falls on the upward side of my view. So the blac... | Not supported with pagination yet | ||||
0a35a578b092cb3a432cf34b2b07e9dadfa40c6e2b6864b07f3a05e15579a0eb | spatial/undirectional_5k | train | Consider the real-world 3D locations of the objects. If I stand at the kitchen sink's position and look at a point roughly between the gray toaster and blue utensil, where does the gray toaster appear to be? | A: back, B: right, C: front, D: left | B | In the kitchen scene, I can identify the sink at the center-left, the gray toaster on the right side of the counter, and the blue utensil holder between them, but raw RGB alone can make the exact object extents and reference points a bit ambiguous. To localize the relevant instances precisely before judging direction f... | In the masked third-person view, I add a red mask to the kitchen sink basin area and a blue mask to the gray toaster, which makes their positions on the countertop much clearer. The utensil holder sits between the sink and toaster in the original scene, so the point roughly between the gray toaster and blue utensil is ... | The directional support view from the sink position clarifies the forward axis by centering the look direction toward the midpoint between the toaster and utensil holder. In that setup, the toaster falls on the right-hand side of the viewing direction rather than the left. Therefore, from the sink while looking at that... | Not supported with pagination yet | ||||
65fe2064ae6a5248f7f7fe6dbd566a3e84ba628a69d0c2e5cac5409c93123529 | spatial/directional_cam_8k | train | Consider the real-world 3D locations and orientations of the objects. Which side of the gray zebra is facing the camera? | A. back-left | B. back-right | C. right | D. front | C | In the original scene, I can clearly see a grazing zebra in profile, but to localize the exact queried animal without any ambiguity, I will generate a third-person view with a colored instance mask. That will isolate the gray zebra from the grassy background and let me anchor the later orientation analysis to the corre... | In the masked third-person view, I add a red mask to the target object, and it cleanly covers the entire zebra from torso to head and legs. This confirms the gray zebra is the only object I need to analyze. To determine which side of the zebra is exposed to the camera rather than just where it sits in the frame, I will... | In the orientation-overlay view, the XYZ axes are anchored on the zebra's body. The horizontal axis runs along the zebra toward its head, establishing the animal's forward direction, while the other axes define the body's upright and lateral directions. Comparing that body-centered orientation with the camera's viewing... | Not supported with pagination yet | ||||
0130f8409bdea00de8709006ef45638ce6da65ae0e6f6fe45dfd755bc8ad4995 | spatial/directional_cam_8k | train | Consider the real-world 3D locations and orientations of the objects. Which side of the red man is facing the camera? | A. front | B. front-right | C. back-left | D. left | A | In the original camera view, I can identify the red man in a mostly frontal standing pose, but to localize the exact queried object cleanly and avoid any ambiguity from clothing and hair details, I will generate a third-person view with a colored instance mask over the target object. | In the masked third-person view, I add a red mask to the target object, and it clearly covers the whole red man from hat to coat and hands. This confirms the queried figure is the full person centered in the scene. To determine which side of the object is exposed to the camera, I will now generate a view with an orient... | In the orientation-overlay view, the XYZ arrows are anchored on the red man and reveal the object's local axes. The upward axis matches the body's vertical direction, while the forward-facing axis projects outward in alignment with the camera's viewing direction rather than away from it. Combined with the original imag... | Not supported with pagination yet | ||||
dd579309a90acf2cc4e36d6c9949c349ce4b013bea57f551fb389b09dbf09473 | spatial/directional_cam_8k | train | Consider the real-world 3D locations and orientations of the objects. Which side of the black person is facing the camera? | A. right | B. back | C. front | D. left | C | In the original racing scene, the black person is the rider leaning low over the motorcycle. Because the rider overlaps heavily with the bike and the track, I first need to localize the exact target cleanly. To remove any ambiguity about which figure the question refers to, I will generate a third-person view with a co... | In the masked third-person view, I add a red mask to the black person, and it clearly covers the rider's helmet, torso, arms, and legs. This confirms the target is the motorcyclist rather than the motorcycle bodywork. With the target now isolated, the remaining question is the rider's intrinsic facing direction relativ... | In the orientation-overlay view, the XYZ arrows are anchored on the rider, letting me read the rider's body orientation rather than just the silhouette. The forward-facing axis aligns generally toward the camera rather than away from it, while the other axes spread sideways and upward from that same body-centered frame... | Not supported with pagination yet | ||||
5b9151d9d99f0ac8341265a0674bb2ad64d2be23a4e2cfa5573384074d33ce49 | spatial/directional_cam_8k | train | Consider the real-world 3D locations and orientations of the objects. Which side of the orange woman is facing the camera? | A. back-right | B. front-right | C. front | D. right | B | In the original court-side photo, I can identify the tennis player as the target woman, but to localize her precisely and avoid any ambiguity with the racket or background people, I will generate a third-person view with a colored instance mask over the target object. | In the masked third-person view, I add a red mask to the target woman, and the overlay cleanly covers her full body from head to shoes, confirming she is the object whose facing side I need to judge. With the target now isolated, the remaining question is her intrinsic orientation rather than just her silhouette, so I ... | In the orientation-overlay view, the axis at her torso shows the upright direction and, more importantly, the horizontal body axes that indicate how she is turned relative to the camera. The forward-facing direction is not straight toward us and not fully side-on; instead it is angled so that her front is partially vis... | Not supported with pagination yet | ||||
9d3f88c53e1cf1dfc49d39cc9726f2333e93aef79a882df16d912c9588f267da | spatial/directional_cam_8k | train | Consider the real-world 3D locations and orientations of the objects. Which side of the black person is facing the camera? | A. front | B. back-right | C. right | D. left | D | In the snowy slope scene, the target is the dark-clothed snowboarder near the center-right. To localize the person precisely and avoid mixing them up with the surrounding trees and board, I will generate a third-person view with a colored instance mask placed on the target person. | In the masked third-person view, I add a red mask to the target snowboarder, which cleanly isolates the black person from the snow, board, and nearby trees. This confirms exactly which figure I need to analyze. To determine which side of this person is facing the camera, I will next generate a view with an orientation ... | In the orientation-overlay view, the axis is anchored on the snowboarder: the blue arrow points upward, and the horizontal axis extends toward the left side of the image from the person's body. Comparing that body-fixed axis with the camera's line of sight, the camera is looking onto the person's left flank rather than... | Not supported with pagination yet |
End of preview. Expand in Data Studio
Instruct-181K
Instruct-181K is the Stage 2 data of PRIMER. It trains ThinkMorph-SSL-Instruct, starting from ThinkMorph-SSL-Base. Each trace has a question, input images, textual reasoning with intermediate images, and an answer.
- Images: the intermediate images are targets built from the original images: localized regions, depth, geometry, reconstructions, paired views and calibrated objects.
- Text: teacher models write the text; generation and filtering depend on the task.
Counts.
- The paper counts 181,291 traces, the rows of the 21 training folders. This release has 180,490 of them: 142,333 with images and 38,157 text only.
- The difference of 801 rows never reached the model in training:
- 800 val rows (
depth_10k,egoexo_5k) have no reasoning image, and Stage 2 drops such rows; - 1 row references a zero-byte image, and the loader skipped it.
- 800 val rows (
- Stage 2 also trained on the val files, so their rows are part of
train.jsonl(source_split=val).
Composition
| Subset | Rows with images | Text-only rows | Sources |
|---|---|---|---|
spatial/behind_me_1k |
1,324 | — | MindCube |
spatial/behind_variants_286 |
284 | — | MindCube |
spatial/directional_query_2k |
2,594 | — | MindCube |
spatial/movement_direction_1k |
1,250 | — | MindCube |
spatial/positioned_direction_423 |
423 | — | MindCube |
spatial/turn_degrees_204 |
204 | — | MindCube |
spatial/turn_move_closer_2k |
2,576 | — | MindCube |
spatial/multiview_58k |
24,247 | 33,631 | Matterport3D (text only), Realsee3D, 360+x |
spatial/depth_10k |
9,575 | — | RefSpatial |
spatial/counting_5k |
5,569 | — | LVIS v1 val (COCO images) |
spatial/rotation_10k |
10,811 | — | MVImgNet |
spatial/perspective_6k |
6,512 | — | SAT |
spatial/directional_1k |
1,162 | — | COCO images (sam3d engine) |
spatial/undirectional_5k |
4,639 | — | COCO images (sam3d engine) |
spatial/directional_cam_8k |
8,063 | — | COCO images (Orient-Anything) |
spatial/jigsaw_20k |
20,000 | — | Pretext-Spatial-50K pools |
perception/jigsaw_20k |
19,935 | — | jigsaw_20k_v1 pools |
perception/visual_search_bbox_15k |
14,686 | — | Visual-CoT and crop pools |
perception/visual_search_crop_7.5k |
7,479 | — | crop pools |
perception/count_1k |
1,000 | — | synthetic counting scenes |
spatial/egoexo_5k |
0 | 4,526 | Ego-Exo4D (text only) |
| Total | 142,333 | 38,157 |
| Functional group (paper) | Pipelines | Paper traces |
|---|---|---|
| L1: visible evidence | visual search, jigsaw QA | 62,100 |
| L2: spatial structure | counting, depth reasoning | 16,544 |
| L3: scene reconstruction | panorama, bird's-eye view, alternate view, object rotation, ego-exo transfer | 102,647 |
| Total | 181,291 |
Files
| Path | What it is |
|---|---|
shards/SHARDS.json, shards/data-*.tar |
The release with images: 142,333 train rows and 286,889 images (157.0 GB), in 148 deterministic tar shards |
text_only/ |
38,157 rows whose images come from Matterport3D or Ego-Exo4D. Their images are not distributed here; see Rows without images |
preview/*.parquet |
The Dataset Viewer preview: a few rows per subset, with images downscaled to at most 768 px |
Download and restore
python scripts/download_release.py shards/SHARDS.json \
--base-url https://huggingface.co/datasets/ThinkMorph/Instruct-181K/resolve/<COMMIT>/shards --output shards
python scripts/shard_data_release.py restore shards/SHARDS.json shards DATA_ROOT/primer-instruct-181k
(hf download ThinkMorph/Instruct-181K --repo-type dataset --include "shards/*" works too.)
download_release.pychecks each shard's size and SHA256.restorechecks every shard before extracting it, then checks every file against theMANIFEST.jsoninside the shards.- A restored release holds
train.jsonl(andval.jsonlif present),images/andMANIFEST.json, which is the layout the BAGEL loader reads. - The scripts are in the PRIMER code release.
Check. Before upload, the restored release went through the BAGEL loader, which decoded every image: 142,333 usable rows and 286,889 images, result PASS.
Row format
{"id": "<sha256>", "question": "...", "options": "...", "images": ["images/<sha256>.jpg"],
"reasoning_thought_0": "...", "reasoning_image_0": "images/<sha256>.jpg",
"reasoning_thought_1": "...", "answer": "...", "subset": "<source subset>", "source_split": "train"}
- The supervision text (
question,options,answer/gt_answer, everyreasoning_thought_N) and the image bytes are exactly as in the training copy. - Image paths are content-hashed relative paths, and
idis a public id. subsetnames the storage subset a row came from, andsource_splitthe split it had there.- Source metadata fields are removed, because they contain internal storage paths.
Rows without images
text_only/ holds the rows whose images come from Matterport3D or Ego-Exo4D. Their terms do not allow
us to redistribute the images. Obtain the source data yourself and rebuild the images.
text_only/train.jsonl(andval.jsonl): the rows, normalized like the main release (same public ids), withsubsetandsource_split. The image fields keep the original file names.text_only/images.jsonl: one record per original image (subset,split,name). Each record gives the byte size and MD5 of the exact file used in training, the upstream dataset and ids, the role and a derivation sentence.text_only/provenance.jsonl: per row, the reconstruction parameters (for example yaw angles of paired views, jigsaw grid and tile permutation, original image id and size, take name and frame rank).
To get the source data:
- Matterport3D: sign the Matterport3D Terms of Use. Get the form from https://niessner.github.io/Matterport/ and send it to matterport3d@googlegroups.com. The rows name the Matterport3D panoramas they use.
- Ego-Exo4D: obtain the dataset under its license (https://ego-exo4d-data.org/). The
frames can be regenerated with
data/builders/egoexo/export_take_ninepack.pyfrom the PRIMER code release; its README gives the exact parameters. An image name gives the take, ego or exo, and the frame rank.
A rebuilt image may differ in bytes even when it shows the same content. Compare it with the recorded MD5.
Source terms
Each source keeps its own terms. The image release has no blanket license. Matterport3D and Ego-Exo4D rows are text only.
Evaluation proximity
- VStar, MMSI and MindCube are close to the visual-search, panorama and viewpoint supervision.
- HRBench and MME-RW-L are nearby perception tasks.
- No evaluation questions or official templates were used to build the data.
Provenance
- Training copy. Training read an internal storage copy that is no longer reachable.
- This release. It is built from the team's archived copy of the same training folders.
- What matches. Row counts agree with the recorded counts of the copy training read.
- What could not be checked. Byte identity with the training copy.
- Downloads last month
- -
Models trained or fine-tuned on ThinkMorph/Instruct-181K