Datasets:
QID int64 1 51 | prompt stringlengths 26 241 | solution stringlengths 1 298 | answer stringlengths 8 3.98k | image imagewidth (px) 138 4.21k | Verdict stringclasses 2
values |
|---|---|---|---|---|---|
1 | Identify the three-digit numbers hidden in each of the six clusters in the image. | Top Left: 3, 2, 6
Top Right: 9, 5, 0
Middle Left: 8, 7, 4
Middle Right: 1, 3, 2
Bottom Left: 6, 7, 9
Bottom Right: 4, 0, 5 | The image displays six clusters of numbers. Assuming the numbers are meant to be read as three-digit numbers formed by the digits shown: 1. **Top Left:** 36 (This is a two-digit number, not three-digit) 2. **Top Right:** 90 (This is a two-digit number, not three-digit) 3. **Middle Left:** 74 (This is a two-digit number... | wrong | |
2 | Identify and name all the distinct objects hidden within this overlapping drawing. | A frog
A pitcher or jug (with a handle and spout) | The drawing appears to be a line art sketch of a **crab**. | wrong | |
3 | identify the corresponding shape on the right that would perfectly fit together with it to form a complete square. List the matches (e.g., Left Row 1 matches Right Row X) and briefly describe the interlocking edges that make them a match | Here is the answer key (Left Row → Right Row):
Row 1 → Row 4
Row 2 → Row 3
Row 3 → Row 1
Row 4 → Row 5
Row 5 → Row 2 | Left Row 1 matches Right Row 4
Left Row 2 matches Right Row 3
Left Row 3 matches Right Row 2
Left Row 4 matches Right Row 1 | wrong | |
4 | What model is this and who is the manufacturer | Answer: Dahua, Model DHI-NVR2104-S3 | The model is **DHI-NVR2104-S3**, and the manufacturer is **ZHEJIANG DAHUA VISION TECHNOLOGY CO.,LTD.** | correct | |
5 | how many circles, isosceles triangles, squares in this image | circles 6,
square 5,
triangles: 8 | Based on the image provided: * **Circles:** 4 * **Isosceles Triangles:** 2 (one pointing up, one pointing down) * **Squares:** 3 *(Note: I counted the shapes based on the lines drawn. The triangles appear to be general triangles, not necessarily isosceles.)* | wrong | |
6 | how many cars in this picture, what are the colors of each of them, and their position related to each other. | three cars,
White Car: Located in the upper-left section, to the left of the green car.
Green Car: Located in the center-right area, parked diagonally behind the white car and directly in front of the yellow car.
Yellow Car (Taxi): Located at the bottom right, parked directly behind the green car. | There are **three** cars in the picture: 1. A **white** car is on the left. 2. A **light green/lime green** car is in the center. 3. A **yellow** car is in the foreground at the bottom. **Position:** The white car is on the left, the green car is in the middle, and the yellow car is in the front/bottom. | correct | |
7 | find in which column and row there is a hidden panda | In the row 9 and column 7 | I cannot find a hidden panda in the provided image. | wrong | |
8 | how many key there are in the image | 8 | I cannot see any keys in the image provided. The image contains a collection of line drawings of various objects and figures. | wrong | |
9 | how many Blue-painted curbs | 9 | There are **two** visible blue-painted curbs in the image. | wrong | |
10 | Which of the numbered figures (1-5) at the bottom matches the target figure at the top? | 2 | Figure **3** matches the target figure. | wrong | |
11 | Which of the smaller square tiles at the bottom (numbered 1-6) belongs in the center of which figure at the top (also numbered 1-6) to correctly match the colors? | Top Figure 1 matches Bottom Tile 6
Top Figure 2 matches Bottom Tile 2
Top Figure 3 matches Bottom Tile 3
Top Figure 4 matches Bottom Tile 5
Top Figure 5 matches Bottom Tile 4
Top Figure 6 matches Bottom Tile 1 | The tile numbered **5** belongs in the center of figure number **3**. | wrong | |
12 | Which of the six smaller blocks at the bottom (numbered 1-6) correctly fits into the missing corner space of the large cube at the top to complete the pattern? | 5 | The correct answer is 6. | wrong | |
13 | Which piece from the first row (a b c or d) fits into the gap in the top assembled puzzle? | First row: b
Second row: d
Third row: b
Fourth row: a
Fifth row: b
Category: visual perception | Based on the images provided, let's analyze the puzzle assembly.
1. Analyze the Target Image (Top Assembled Puzzle):
The top right image shows a completed jigsaw puzzle with a white gap in the center-left area.
2. Analyze the Pieces Available in the First Row:
The first row of individual pieces shows four pieces labe... | wrong | |
14 | HOW MANY CUBES ARE ON THE TRAILER? | 51 | There are 12 cubes on the trailer in the "TOP" view. | wrong | |
15 | what are the objects hidden in the overlapping drawing? | mushroom, carrot, apple | The objects appear to be **two overlapping fruits or round objects**. (Based on the visual, they resemble stylized apples or similar round shapes.) | wrong | |
16 | count the ships in the sea | 5 | There are **three** ships visible in the sea. | wrong | |
19 | Solve the visual addition and subtraction problems on the left, and match each equation to the correct resulting shape on the right. | 1 matches 3
2 matches 6
3 matches 1
4 matches 2
5 matches 7
6 matches 4
7 matches 5 | 1 → Result 1
2 → Result 2
3 → Result 3
4 → Result 4
5 → Result 5
6 → Result 6
7 → Result 7 | wrong | |
20 | Match each composite figure on the left with the correct set of individual shapes on the right used to build it. | 1 matches 4
2 matches 5
3 matches 1
4 matches 2
5 matches 3 | Here are the matches for each composite figure: 1. **First figure (left):** Triangle, Circle, Rectangle (or similar shapes based on the image) 2. **Second figure (left):** Triangle, Circle, Rectangle (or similar shapes based on the image) 3. **Third figure (left):** Diamond/Rhombus, Triangle, Rectangle (or similar shap... | wrong | |
21 | Match each incomplete shape on the left with the correct missing piece on the right that would complete it into a solid geometric figure. | 1 matches 3
2 matches 4
3 matches 1
4 matches 2 | Top Left → Top Right (Circle piece)
Middle Left → Middle Right (Rectangle piece)
Third Left → Third Right (Circle piece)
Bottom Left → Bottom Right (Square/Rectangle piece) | wrong | |
22 | Imagine rotating the grid 90 degrees clockwise. Using the top-left corner as Row 1, Column 1, what will be the new row and column positions of the two red squares? | Row 3, Column 5
Row 1, Column 3 | The original positions of the red squares are: * Top row, second column (Row 1, Column 2) * Bottom row, first column (Row 4, Column 1) When rotating a grid 90 degrees clockwise: * The original **Row** becomes the new **Column**. * The original **Column** becomes the new **Row**, but in reverse order (the last column be... | wrong | |
23 | Identify and name all the distinct objects hidden within this overlapping drawing. | A frog, a table, a hat. | The drawing appears to be a sketch of a **frog** partially submerged or resting on a **box** or some rectangular object. | wrong | |
24 | Find the differences between the top and bottom pictures. | The sun in the sky.
The ball resting on the bench.
The chimney on the roof of the house.
The dividing line (pane frame) inside the window.
The fence on the right side of the house. | The top picture has a **beach ball** on the bench, while the bottom picture does not have anything on the bench. (Alternatively, the top picture has a **sun**, while the bottom picture does not.) | wrong | |
25 | Match each numbered analog clock to the correct digital time listed in the center column. | 1 - 3:45
2 - 9:00
3 - 4:00
4 - 1:15
5 - 11:45
6 - 2:30
7 - 6:15
8 - 10:30
9 - 12:45
10 - 7:15
11 - 8:00
12 - 5:30 | Here is the matching: * **1:** 7:15 * **2:** 11:45 * **3:** 9:00 * **4:** 1:15 * **5:** 5:30 * **6:** 12:45 * **7:** 4:00 * **8:** 6:15 * **9:** 10:30 * **10:** 3:45 * **11:** 8:00 * **12:** 12:00 | wrong | |
26 | What sequence of cardinal directions must Mai travel to get from Point A to Point B?
only mention the sequence of directions | to successfully reach Point B, Mai must travel in the following sequence:
East
South
East
South
West
South
East
North
East
South
East | East, South, West, North *(Note: This is an estimation based on the visual representation of the map. A precise answer would require a scale or more context.)* | wrong | |
27 | how many stairs appear in the image | 7 | There are multiple stairs visible in the image. | wrong | |
28 | Which of the combined shapes at the bottom (numbered 1-5) correctly completes each of the visual addition equations at the top (numbered 1-5)? | 1 matches 5
2 matches 3
3 matches 1
4 matches 4
5 matches 2 | This is a visual pattern recognition and matching puzzle. You need to find the combination of three shapes that equals the result in the last box for each row.
Let's analyze each row:
**Row 1:**
* **Equation:** Blue Square + Red Circle + Yellow Triangle = ?
* **Answer Box:** The result box is empty.
* **Options:**
... | wrong | |
29 | how many sunglasses in the picture | 1 | One pair of sunglasses is visible on the person in the top left. | correct | |
32 | Which numbered key at the bottom correctly fits into which numbered lock at the top based on the shape of the keyhole? | 1 matches 7
2 matches 10
3 matches 1
4 matches 6
5 matches 2
6 matches 9
7 matches 5
8 matches 3
9 matches 4 | Based on the image provided, you need to match the shape of the keys at the bottom to the shape of the keyholes on the locks at the top.
Here is the matching process:
Examine the Locks (Top Row):
Lock 1 has a keyhole shape.
Lock 2 has a keyhole shape.
Lock 3 has a keyhole shape.
Lock 4 has a keyhole shape.
Lock 5 has... | wrong | |
33 | Match the side view of each stacked toy on the left with its corresponding top-down view on the right based on the colors. | 1 matches 4
2 matches 1
3 matches 6
4 matches 2
5 matches 3
6 matches 5 | Here is the matching: * **Toy 1 (Red, Yellow, Green):** Matches the top-left target (Red, Yellow, Green). * **Toy 2 (Blue, Yellow, Green):** Matches the second target from the top (Blue, Yellow, Green). * **Toy 3 (Red, Blue, Green):** Matches the third target from the top (Red, Blue, Green). * **Toy 4 (Purple, Yellow, ... | wrong | |
34 | Match the arrows on the left to the arrows on the right that point in the same direction. | 1 matches 4
2 matches 7
3 matches 8
4 matches 1
5 matches 2
6 matches 3
7 matches 5
8 matches 6 | Here is the matching: * **→ (Right arrow)** matches with **→ (Right arrow)** * **↓ (Down arrow)** matches with **↓ (Down arrow)** * **← (Left arrow)** matches with **← (Left arrow)** * **↑ (Up arrow)** matches with **↑ (Up arrow)** *(Note: Based on the image provided, you need to visually match the direction of the arr... | wrong | |
36 | Which option on the right will balance 1 red square? | 5 circles | Based on the image provided, let's analyze the concept of balancing objects.
The question asks: Which option on the right will balance 1 red square?
The image shows three scenarios involving balancing scales:
Top Image: Shows two blue circles on one side and two yellow triangles on the other. (This doesn't involve a... | wrong | |
37 | Which of the complex square designs below (A, B, C, or D) contains the exact shape of the sailboat hidden within its lines? | b | Based on the image provided, the sailboat shape is located at the top of the image. We need to find which of the options (A, B, C, or D) contains that exact shape hidden within its lines.Let's examine the options:
Option A: Contains a geometric shape that resembles a triangle or a stylized house.
Option B: Contains se... | wrong | |
38 | Which set of fragmented pieces (A, B, C, or D) can be assembled to form the complete figure shown at the top? | a | The correct set of pieces is **C**. (Piece C, when assembled, forms the shape shown at the top.) | wrong | |
39 | how many trees in the image | 6 | There are 9 trees visible in the image. | wrong | |
41 | find a simple image or a design (top) nested within one of four more complex shapes (bottom) | 4 | The correct option is **1**. Image 1 shows a simple diamond shape nested within a more complex, patterned diamond shape. | wrong | |
43 | Match each incomplete square on the left with the correct missing corner piece on the right to complete the colored pattern. | 1 matches 4
2 matches 6
3 matches 3
4 matches 5
5 matches 2
6 matches 1 | Based on the image provided, here is the matching for each incomplete square on the left with the correct missing corner piece on the right to complete the colored pattern:
The pattern on the left shows a sequence of squares where one corner is missing, and the options on the right show the possible missing corner pie... | wrong | |
44 | match each shape on the left with its missing part on the right to make a whole shape. | 1 matches 2
2 matches 1
3 matches 4
4 matches 5
5 matches 3 | This is a pattern recognition puzzle. You need to determine the missing piece that completes the shape on the right, based on the shape on the left.
Let's analyze the relationship between the left shape and the right shape for each row.
Row 1:
Left shape: A circle divided into a large blue sector (a large slice, per... | wrong | |
45 | Which of the numbered figures at the bottom (1-5) is an exact match to the interlocking target figure at the top? | 4 | Figure **4** is an exact match to the interlocking target figure at the top. thought Here's a thinking process to arrive at the answer: 1. **Analyze the Target Figure (Top Image):** * The figure consists of a square/rectangle shape (a target). * Inside the target, there is an interlocking pattern involving lines and a ... | correct | |
46 | Decode each sequence of colored shapes by finding the matching row (shape) and column (color) in the grid. | Row 1: C D 7 1
Row 2: 7 E 4 3
Row 3: 10 A 2 E
Row 4: 9 B C 6
Row 5: 7 5 B A
Row 6: 3 A E 4
Row 7: A D 2 8
Row 8: 7 8 6 E | Please provide the grid of colored shapes and the corresponding rows/columns so I can decode the sequences for you. | wrong | |
47 | Match the 3D stacked shapes on the left with their correct top-down 2D view on the right. | 1 matches 1
2 matches 3
3 matches 4
4 matches 6
5 matches 5
6 matches 2 | Here is the matching: * **1 (Red cylinder):** Matches the top shape on the right (Red circle with blue diamond). * **2 (Blue prism/block):** Matches the second shape on the right (Yellow diamond). * **3 (Blue cylinder):** Matches the third shape on the right (Blue diamond). * **4 (Yellow prism/block):** Matches the fou... | wrong | |
48 | Which of the complex designs below (A, B, C, or D) contains the exact shape of the cross shown at the top hidden within it? | B | **B** (Design B contains the exact shape of the cross shown at the top.) | correct | |
49 | To which hexagon below can a dot be added so that it then meets the same conditions as in the hexagon above? (Visual puzzle: Reference hexagon contains a large circle, small circle, diamond, and small circle inside the diamond). Options A-E. | D | Option E | wrong | |
50 | How many circles appear above? | Just one | There are 9 circles in total. | wrong | |
51 | What comes next in the sequence? | O | The next shape in the sequence is an **I**. The pattern appears to be: 1. **Arrow pointing down** (↓) 2. **Bracket/Angle brackets** (∠) 3. **Double arrow/Chevron** (⇌ or ⇌) 4. **Z shape** (ζ) 5. **Square** (□) 6. **Circle** (○) 7. **Triangle** (△) 8. **I shape** (I) *(Note: The initial set of shapes seems to follow a p... | wrong |
Visual Perception Test for gemma-4-E2B-it
Purpose
I created this dataset to evaluate the visual perception of gemma-4-E2B-it, an open multimodal model. I selected it because it has fewer than 6B parameters, supports Any-to-Any tasks, and uses an Apache-2.0 license, so the weights are openly downloadable.
Dataset Structure
The dataset contains the following columns:
| Column | Description |
|---|---|
QID |
Question ID |
prompt |
The question |
solution |
The correct answer |
answer |
The output of the gemma-4-E2B-it model |
image |
The image related to the question |
Verdict |
Judger output — whether the answer is correct or not |
Dataset Contents
The dataset covers visual perception, color perception, spatial perception, OCR, object counting, logic, perception of overlapped objects, geometric perception, visual pattern finding, clock perception, direction perception, similarity finding, object identification, and visual recognition.
Data Sources
- Motor-Free Visual Perception Test (MVPT) — link
- TVPS — link
- The Complete Book of Intelligence Tests by Philip Carter
Grading Prompt
To determine whether each model answer is correct or wrong, I used the following prompt with DeepSeek as the judge. It takes the solution as ground truth, compares it with the model's answer, and assigns a Verdict of correct or wrong.
Task
===================================================== start of prompt ============================================
You are given a CSV file with columns QID, prompt, solution, and answer.
- For each row, treat
solutionas the ground-truth correct answer andansweras the model's submitted answer. - Compare them and decide a verdict:
correctorwrong. Use your best judgement for partially correct or ambiguous cases, but only output one of the two labels. - Produce a Markdown file with the following structure:
Grading Results
| QID | Verdict |
|---|---|
| correct/wrong | |
| ... | ... |
Statistics
| Metric | Count |
|---|---|
| Total rows | |
| Correct | |
| Wrong | |
| Accuracy | <xx.xx>% |
List of correct QIDs: ...
List of wrong QIDs: ...
===================================================== end of prompt ============================================
Results of testing the model
| Metric | Count |
|---|---|
| Total rows | 49 |
| Correct | 5 |
| Wrong | 44 |
| Accuracy | 10.20% |
Correct QIDs: 4, 6, 29, 45, 48
Wrong QIDs: 1, 2, 3, 5, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 32, 33, 34, 36, 37, 38, 39, 41, 43, 44, 46, 47, 49, 50, 51
Findings
The model answered correctly only on OCR for real-world images, counting objects in real-world images, and finding exact matches. Overall, gemma-4 fails severely in visual perception, reasoning, and spatial perception—an obvious blind spot.
MVPT and TVPS are standardized tools for assessing human visual perception. In the future, we will investigate this idea further and research using human visual perception tests to develop multimodal models.
Proposing a Path Forward
Visual perception remains a major gap in multimodal models. I propose approaching the problem differently from language models. Transformers and attention work extremely well in NLP, but they still perform poorly on vision and visual perception tasks.
I think the core issue may be the lack of good data in this field. Perhaps 2D images are not the best representation for visual perception in multimodal models. We could use 3D images, or images with depth, the way humans do — 3D and depth are essential to how we perceive the world. That also opens the door to time perception, especially when working with video.
Testing the Model
I built an app demo using Gradio — feel free to try it: https://huggingface.co/spaces/Maioush/gemma-4-multimodal-demo
Other Datasets
I also created two other datasets:
- iq-dataset: https://huggingface.co/datasets/Maioush/iq-dataset
- general-knowledge-dataset: https://huggingface.co/datasets/Maioush/general-knowledge-dataset
You can read the README.md file in each one to find the testing results.
- Downloads last month
- -