ruotian/SelectGround-8B
Image-Text-to-Text • Updated
Datasets:
id stringlengths 18 18 | image stringlengths 75 75 | image_width int64 196 3.75k | image_height int64 224 1.93k | image_sha256 stringlengths 64 64 | instruction stringlengths 4 506 | response stringlengths 6 9 | source dict | target_bbox listlengths 4 4 | distractor_bbox listlengths 4 4 | candidate_bboxes listlengths 3 736 |
|---|---|---|---|---|---|---|---|---|---|---|
train_pairs-000001 | images/fb02a8b7ac7339013370b5a30ff6b1eee72f709ac0e3e3b0904fc7fc0a8e0785.jpg | 1,288 | 728 | fb02a8b7ac7339013370b5a30ff6b1eee72f709ac0e3e3b0904fc7fc0a8e0785 | Enable description display | [199,485] | {
"dataset": "mlfoundations/Click-100k",
"split": "train",
"index": 98334
} | [
250,
347,
262,
359
] | [
271,
347,
329,
361
] | [
[
653,
9,
715,
23
],
[
721,
9,
763,
21
],
[
17,
29,
45,
41
],
[
61,
31,
97,
45
],
[
115,
31,
175,
43
],
[
193,
31,
247,
45
],
[
265,
31,
305,
45
],
[
323,
31,
355,
... |
train_pairs-000002 | images/9d44c847d1c649004980a842af69207da522bbf4b28458f2e8f8b8f625bbb61d.png | 420 | 252 | 9d44c847d1c649004980a842af69207da522bbf4b28458f2e8f8b8f625bbb61d | Using the image of this webpage, can you determine the coordinates of the element I describe (with point)? This element provides a way to navigate to the main homepage of the website, allowing users to access the website's main content and features. | [881,36] | {
"dataset": "mlfoundations/Click-100k",
"split": "train",
"index": 35364
} | [
352,
0,
388,
18
] | [
313.2265625,
0,
351.3150329589844,
16.53608512878418
] | [
[
334,
50,
384,
58
],
[
26.55591583251953,
0.20206600427627563,
43.467472076416016,
15.809082984924316
],
[
67.89698028564453,
0.7049435377120972,
79.58617401123047,
15.310229301452637
],
[
55.66249465942383,
1.095736026763916,
67.1207962036132... |
train_pairs-000003 | images/178dd89d8410e4c5aea8e1e7deba19718dbf25572ebd1d919cf9de2e9c7020b0.jpg | 616 | 476 | 178dd89d8410e4c5aea8e1e7deba19718dbf25572ebd1d919cf9de2e9c7020b0 | Select to view the timeline of the COVID-19 pandemic | [442,819] | {
"dataset": "mlfoundations/Click-100k",
"split": "train",
"index": 35585
} | [
256,
385,
289,
395
] | [
261,
367,
357,
383
] | [
[
69,
9,
135,
25
],
[
181,
11,
259,
25
],
[
31,
65,
169,
79
],
[
141,
95,
185,
109
],
[
203,
95,
317,
109
],
[
335,
97,
379,
109
],
[
403,
97,
443,
111
],
[
451,
97,
... |
train_pairs-000004 | images/decb38e3e91b4d668d300d06e63f0a1a4a94bb193c0fb3ed2f42632415e72c37.jpg | 1,568 | 896 | decb38e3e91b4d668d300d06e63f0a1a4a94bb193c0fb3ed2f42632415e72c37 | Toggle bold weight on selected text | [637,130] | {
"dataset": "mlfoundations/Click-100k",
"split": "train",
"index": 28530
} | [
986,
104,
1011,
129
] | [
940,
840,
1246,
866
] | [
[
31,
7,
266,
32
],
[
283,
9,
456,
29
],
[
471,
9,
606,
33
],
[
612,
8,
724,
32
],
[
21,
47,
47,
63
],
[
58,
42,
92,
66
],
[
103,
45,
141,
63
],
[
153,
47,
197,
... |
train_pairs-000005 | images/bfb40ea3ec3ea786648272220a91ea75106cb8aeb25209d9e0fcabe0d38fedad.jpg | 1,288 | 728 | bfb40ea3ec3ea786648272220a91ea75106cb8aeb25209d9e0fcabe0d38fedad | Access window options | [218,50] | {
"dataset": "mlfoundations/Click-100k",
"split": "train",
"index": 100959
} | [
258,
29,
303,
44
] | [
102.16974639892578,
22.21894073486328,
133.41708374023438,
49.585506439208984
] | [
[
99,
5,
167,
17
],
[
173,
3,
215,
19
],
[
221,
3,
285,
19
],
[
349,
3,
389,
19
],
[
31,
13,
73,
25
],
[
77,
31,
99,
47
],
[
107,
31,
131,
47
],
[
137,
31,
169,
... |
train_pairs-000006 | images/10946db27cc7b571e1db3031ed85a726d77eabf7456550666a5eb385fdd1b9f4.png | 224 | 364 | 10946db27cc7b571e1db3031ed85a726d77eabf7456550666a5eb385fdd1b9f4 | In the provided web page image, I'll describe a specific element. Identify their locations, please (with point). This element serves as a navigational link, facilitating user navigation back to the home page of the blog. | [174,253] | {
"dataset": "mlfoundations/Click-100k",
"split": "train",
"index": 4015
} | [
15,
84,
63,
100
] | [
1.3195924758911133,
42.447998046875,
222.98265075683594,
77.88164520263672
] | [
[
59,
55,
167,
71
],
[
0.12778997421264648,
23.888046264648438,
80.43889617919922,
43.309967041015625
],
[
3.9994394779205322,
180.299072265625,
103.7370376586914,
197.39353942871094
],
[
4.512446880340576,
164.39569091796875,
69.15963745117188... |
train_pairs-000007 | images/4715f1193cc3199c7e8fc08dfd6a0d2fdf58540d9755d92a9614ec92ceea1b5f.png | 616 | 364 | 4715f1193cc3199c7e8fc08dfd6a0d2fdf58540d9755d92a9614ec92ceea1b5f | Using the image of this webpage, can you determine the coordinates of the element I describe (with point)? This element provides access to documentation on managing organizations within a Gandi account. | [118,279] | {
"dataset": "mlfoundations/Click-100k",
"split": "train",
"index": 15992
} | [
0,
91,
145,
112
] | [
199.93338012695312,
248.8619384765625,
334.34381103515625,
260.13018798828125
] | [
[
20,
3,
67,
21
],
[
8,
34,
48,
40
],
[
7,
55,
99,
67
],
[
7,
74,
106,
88
],
[
182,
68,
297,
86
],
[
7,
95,
93,
107
],
[
184,
108,
224,
116
],
[
300,
108,
398,
... |
train_pairs-000008 | images/c0314968fcffc4cbf4fa5d837615e7aa1b3e4d2ab6e0b18c89402b4e17474cc0.jpg | 1,288 | 728 | c0314968fcffc4cbf4fa5d837615e7aa1b3e4d2ab6e0b18c89402b4e17474cc0 | Click on "Enhance video" | [774,73] | {
"dataset": "mlfoundations/Click-100k",
"split": "train",
"index": 81168
} | [
977,
40,
1017,
66
] | [
941.5540771484375,
38.0227165222168,
968.8807373046875,
66.75147247314453
] | [
[
15,
7,
97,
27
],
[
158,
4,
224,
28
],
[
726,
4,
766,
28
],
[
826,
4,
932,
30
],
[
323,
185,
393,
205
],
[
429,
185,
507,
205
],
[
542,
182,
660,
206
],
[
293,
251,
... |
train_pairs-000009 | images/d1edfd3e164353484b53c03a3f2b2b7120a6e2e809e8ac3356bae2199d19e83c.png | 1,372 | 756 | d1edfd3e164353484b53c03a3f2b2b7120a6e2e809e8ac3356bae2199d19e83c | To the left of the "path", which ui element is positioned closest horizontally? | [132,46] | {
"dataset": "mlfoundations/Click-100k",
"split": "train",
"index": 6656
} | [
159,
25,
204,
44
] | [
0,
0.6724094152450562,
20.595796585083008,
22.154150009155273
] | [
[
23,
4,
181,
22
],
[
7,
29,
29,
43
],
[
41,
29,
67,
43
],
[
77,
29,
107,
43
],
[
117,
29,
153,
43
],
[
163,
29,
203,
43
],
[
213,
29,
241,
43
],
[
253,
29,
279,
... |
train_pairs-000010 | images/16f446402fedabbb6e55a8f097e2e7ca3c31e709aa8e04463b2b66187e704f96.png | 448 | 252 | 16f446402fedabbb6e55a8f097e2e7ca3c31e709aa8e04463b2b66187e704f96 | Given a screenshot, I will describe a specific element; your task is to predict their locations (with point). This element launches the Gmail application, displaying a welcome tour highlighting its updated features. | [117,964] | {
"dataset": "mlfoundations/Click-100k",
"split": "train",
"index": 21524
} | [
37,
234,
68,
252
] | [
21.496580123901367,
114.99842071533203,
83.92984008789062,
227.8828125
] | [
[
347.57769775390625,
15.465904235839844,
394.22027587890625,
42.268394470214844
],
[
125.2734375,
115.45138549804688,
180.22811889648438,
176.30758666992188
],
[
54.167152404785156,
15.446685791015625,
80.35980987548828,
42.634666442871094
],
[
232.06... |
train_pairs-000011 | images/636417cc0ee928f5b8166b30791890c1bf24bb7bcc8bfc308e7b0697393f4d74.png | 224 | 364 | 636417cc0ee928f5b8166b30791890c1bf24bb7bcc8bfc308e7b0697393f4d74 | I'll provide textual descriptions of the element in this webpage screenshot. Can you find their coordinates (with point)? This element acts as a gateway to the Yahoo platform, directing users to the respective service upon interaction. | [493,54] | {
"dataset": "mlfoundations/Click-100k",
"split": "train",
"index": 46769
} | [
80,
7,
141,
32
] | [
101,
329,
143,
343
] | [
[
87,
13,
137,
29
],
[
43,
75,
183,
91
],
[
97,
193,
129,
205
],
[
63,
269,
103,
283
],
[
115,
271,
163,
283
],
[
73,
305,
131,
317
],
[
101,
329,
143,
343
],
[
78.08595119... |
train_pairs-000012 | images/4e755e423eda5114c6350a582382241b7c7fda9234a538285edcef82e8b3a3ce.jpg | 1,932 | 1,092 | 4e755e423eda5114c6350a582382241b7c7fda9234a538285edcef82e8b3a3ce | Minimize objects view | [918,272] | {
"dataset": "mlfoundations/Click-100k",
"split": "train",
"index": 99697
} | [
1630,
288,
1917,
305
] | [
289.4381408691406,
532.9890747070312,
315.1025085449219,
559.883544921875
] | [
[
23,
5,
105,
21
],
[
113,
5,
231,
19
],
[
17,
31,
41,
45
],
[
65,
31,
93,
45
],
[
113,
31,
147,
45
],
[
161,
31,
217,
45
],
[
233,
31,
267,
45
],
[
283,
31,
359,
... |
train_pairs-000013 | images/eb1fea5520e40090df898343ce6ce4f8b6f0c495a1630a616acbadf92e359eeb.jpg | 2,268 | 1,288 | eb1fea5520e40090df898343ce6ce4f8b6f0c495a1630a616acbadf92e359eeb | Open the Rendering menu for output settings | [396,13] | {
"dataset": "mlfoundations/Click-100k",
"split": "train",
"index": 2493
} | [
857,
2,
938,
32
] | [
119,
9,
165,
25
] | [
[
39,
7,
67,
25
],
[
79,
7,
107,
25
],
[
119,
9,
165,
25
],
[
177,
9,
231,
25
],
[
243,
9,
275,
25
],
[
309,
13,
355,
27
],
[
373,
13,
433,
29
],
[
448,
10,
509,
... |
train_pairs-000014 | images/c189303ae08ae3df142440015f634a1248670dc26fba12bddf189866a463c956.jpg | 1,288 | 728 | c189303ae08ae3df142440015f634a1248670dc26fba12bddf189866a463c956 | Click on "Utah sues federal government - Search - Sleeping" | [166,24] | {
"dataset": "mlfoundations/Click-100k",
"split": "train",
"index": 82274
} | [
195,
0,
233,
35
] | [
488.2406005859375,
5.650662899017334,
504.1792297363281,
29.743911743164062
] | [
[
704,
92,
734,
124
],
[
782,
94,
808,
120
],
[
918,
89,
1068,
116
],
[
0,
157,
121,
173
],
[
148,
155,
209,
175
],
[
237,
157,
289,
173
],
[
295,
157,
345,
173
],
[
495,
... |
train_pairs-000015 | images/ec1488a8740041245a8e39cf80612e51dd0ec14db1c3101449be1186b30065f2.png | 1,932 | 1,092 | ec1488a8740041245a8e39cf80612e51dd0ec14db1c3101449be1186b30065f2 | Vertically below the "latest nba score today", which ui element is positioned closest? | [277,331] | {
"dataset": "mlfoundations/Click-100k",
"split": "train",
"index": 23632
} | [
419,
346,
653,
377
] | [
586.2135009765625,
232.49888610839844,
1535.90087890625,
488.1911926269531
] | [
[
47,
41,
109,
59
],
[
120,
86,
190,
112
],
[
460,
96,
600,
120
],
[
606,
94,
772,
118
],
[
619,
137,
679,
153
],
[
707,
135,
753,
153
],
[
781,
133,
837,
153
],
[
461,
... |
train_pairs-000016 | images/8d4b865b4137860edd16786eca38ad0d28fe19d6cc2c42730859b8400bda2019.png | 2,548 | 1,540 | 8d4b865b4137860edd16786eca38ad0d28fe19d6cc2c42730859b8400bda2019 | Insert the current time into the document | [616,139] | {
"dataset": "mlfoundations/Click-100k",
"split": "train",
"index": 11144
} | [
1553,
200,
1588,
228
] | [
2127,
133,
2302,
160
] | [
[
90,
18,
166,
46
],
[
194,
18,
238,
44
],
[
268,
18,
316,
44
],
[
344,
18,
400,
44
],
[
432,
20,
468,
44
],
[
498,
18,
606,
44
],
[
636,
20,
694,
44
],
[
724,
20,
... |
train_pairs-000017 | images/cb2c47a6f50202e1afff12e94965f1e16a9619ac77a9fa10c5d4b1cbf7873d5b.jpg | 980 | 560 | cb2c47a6f50202e1afff12e94965f1e16a9619ac77a9fa10c5d4b1cbf7873d5b | Select the area to navigate to the clojureD 2022 event page | [363,530] | {
"dataset": "mlfoundations/Click-100k",
"split": "train",
"index": 2689
} | [
314,
287,
398,
307
] | [
313.1960144042969,
267.4885559082031,
457.0396423339844,
287.6794738769531
] | [
[
129,
15,
234,
60
],
[
355,
31,
413,
45
],
[
431,
31,
495,
45
],
[
513,
31,
537,
45
],
[
553,
31,
609,
45
],
[
627,
31,
671,
45
],
[
687,
31,
763,
45
],
[
779,
31,
... |
train_pairs-000018 | images/d30e5714f4bfe9ec306e363b73c63b5481b3eeb8a9e769a56f0e76e7adf4cad9.jpg | 2,660 | 1,484 | d30e5714f4bfe9ec306e363b73c63b5481b3eeb8a9e769a56f0e76e7adf4cad9 | Generate AI insights | [762,151] | {
"dataset": "mlfoundations/Click-100k",
"split": "train",
"index": 99549
} | [
1935,
210,
2120,
237
] | [
1976,
165,
2178,
195
] | [
[
128,
6,
209,
31
],
[
221,
7,
287,
34
],
[
347,
8,
407,
33
],
[
30,
48,
65,
69
],
[
117,
48,
171,
69
],
[
215,
48,
300,
69
],
[
343,
48,
451,
69
],
[
493,
48,
539,... |
End of preview. Expand in Data Studio
ClickContrast
ClickContrast contains 4,292 target–distractor pairs and 4,292 coverage examples from Click-100k, with 6,690 distinct screenshots. It is the training dataset for SelectGround-8B.
| File | Rows | Supervision |
|---|---|---|
| train_pairs | 3,790 | Target coordinate, target region, distractor region, candidate regions |
| train_replay | 3,790 | Target coordinate |
| refinement_pairs | 502 | Target coordinate, target region, distractor region, candidate regions |
| refinement_replay | 502 | Target coordinate |
The initial training stage uses all four files, with a 1:1 pair-to-coverage sampling cycle. Refinement uses only the two refinement files, with two coverage microbatches per paired microbatch. The refinement rows are included in the 8,584-row total.
Main pairs retain targets supported by both MAI-UI-8B and UI-Venus-1.5-8B, together with an incorrect Qwen3-VL-8B prediction on another parsed element outside a centered 2x target neighborhood. Refinement pairs focus on teacher-checked small controls and icons, using wrong-element predictions from a fine-tuned mining model. Coverage examples span 86 Click100k source categories.
Format
Download the entire repository to retain the relative screenshot paths:
hf download ruotian/ClickContrast --repo-type dataset --revision paper --local-dir data/clickcontrast
Each row contains:
- id: stable example identifier.
- image: screenshot path relative to this repository.
- image_width, image_height, image_sha256: original dimensions and file checksum.
- instruction: original grounding instruction.
- response: target center as [x,y], with integer coordinates normalized to 0–1000.
- source: upstream dataset, split, and row index.
Paired rows additionally contain target_bbox, distractor_bbox, and candidate_bboxes, all in original-image pixels with [left,top,right,bottom] order. The JSONL order is part of the frozen training recipe. The manifest records file hashes and counts.
Training
Use the two-stage SFT-plus-selection-loss recipe in the SelectGround code repository. All examples contribute coordinate loss. Only paired examples contribute selection loss. No external parser or teacher is needed to train from this release.
Source terms
Screenshots are a subset of Click-100k and retain their upstream provenance. They are not relicensed as project code. Use is subject to the applicable terms of Click-100k and its constituent sources. The repository's code license does not replace those data terms.
- Downloads last month
- -
Models trained or fine-tuned on ruotian/ClickContrast