HongxinLi commited on
Commit
94290e0
·
verified ·
1 Parent(s): 3317b1e

Add dataset card (README) with schema and field documentation

Browse files
Files changed (1) hide show
  1. README.md +169 -0
README.md ADDED
@@ -0,0 +1,169 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: cc-by-nc-4.0
3
+ task_categories:
4
+ - image-to-text
5
+ - visual-question-answering
6
+ language:
7
+ - en
8
+ tags:
9
+ - gui
10
+ - gui-agent
11
+ - ui-understanding
12
+ - screenshot
13
+ - visual-grounding
14
+ pretty_name: UIPro WebUI
15
+ size_categories:
16
+ - 100K<n<1M
17
+ ---
18
+
19
+ # UIPro-WebUI-Data-v1
20
+
21
+ Part of the **UIPro** GUI-agent training suite (ICCV 2025). This repository packages the
22
+ **WebUI** source into the unified UIPro instruction-tuning format, with coordinates
23
+ normalized to a **[0, 1000]** grid.
24
+
25
+ > Web-page screenshots with icon grounding/referring, element classification, intent grounding and widget listing.
26
+
27
+ ## Dataset at a glance
28
+
29
+ | | |
30
+ | :--- | :--- |
31
+ | **Total samples** | 488,499 |
32
+ | **Valid images** | 327,280 |
33
+ | **Avg. samples / image** | 1.49 |
34
+ | **Coordinate scale** | 0–1000 |
35
+ | **Source dataset** | [WebUI](https://huggingface.co/datasets/biglab/webui-all) |
36
+
37
+ ### Samples by task
38
+
39
+ | Task | Count |
40
+ | :--- | ---: |
41
+ | WidgetList | 160,266 |
42
+ | IconRef | 116,403 |
43
+ | IntentGnd | 110,800 |
44
+ | IconGnd | 59,561 |
45
+ | ElemClass | 41,469 |
46
+
47
+ ## Repository file structure
48
+
49
+ | File | Description |
50
+ | :--- | :--- |
51
+ | `WebUI_scale1000_488k.json` | The dataset: a JSON **list** of 488,499 sample objects (schema below). |
52
+ | `WebUI_scale1000_488k_sample.json` | A small preview slice of the same schema, for quick inspection without downloading everything. |
53
+ | `WebUI_scale1000_488k_images.zip` | All screenshots referenced by the `image` field, preserving the relative paths stored there. |
54
+ | `WebUI_scale1000_488k_info.json` | Full generation report — per-task counts, image statistics, invalid-element breakdown, and the exact processing config. |
55
+
56
+ Unzip `WebUI_scale1000_488k_images.zip` and each sample's `image` path resolves relative to the extraction root.
57
+
58
+ ## Sample schema — every field explained
59
+
60
+ Each element of the main JSON list is one training sample. This dataset's samples use the
61
+ following fields:
62
+
63
+ | Field | Meaning |
64
+ | :--- | :--- |
65
+ | `conversations` | The vision-language dialogue: a list of turns, each `{"from": "human"|"gpt", "value": ...}`. The **human** turn holds the instruction/question and contains the `<image>` placeholder marking where the screenshot is inserted; the **gpt** turn is the ground-truth answer. |
66
+ | `id` | Unique sample identifier, formatted `autogui_<dataset>_<task>_<n>`. The `<task>` segment (e.g. `intentgnd`, `textloc`, `ocr`, `elemgnd`, `elemref`) tells you which task the sample belongs to. |
67
+ | `image` | Path to the screenshot **inside `_images.zip`**, relative to the archive root. Load the image by joining this path with your extraction directory. |
68
+ | `task_attr` | The task's target attribute in plain form — for grounding tasks the referred element's text/instruction; for OCR/referring tasks the queried coordinate string. Useful for filtering or building custom prompts without parsing the conversation. |
69
+ | `unnormalized_box` | Ground-truth bounding box in **original image pixels**, as `[x1, y1, x2, y2]` (top-left, bottom-right). Present when a box is available. Note: the answer in the `gpt` turn is **normalized to 0–1000**, while this field is the raw-pixel box — divide by width/height and multiply by 1000 to reconcile them. |
70
+ | `wxh` | Original screenshot size as the string `"<width>x<height>"` in pixels — use it to convert between the normalized 0–1000 coordinates and raw pixels. |
71
+
72
+ > **Note:** Fields tied to a bounding box (e.g. `unnormalized_box`) are only present on samples
73
+ > that have a box; point-only answers (e.g. some intent-grounding samples) may omit them.
74
+
75
+ ### The `conversations` field in detail
76
+
77
+ `conversations` is a list of turns that a vision-language model consumes directly:
78
+
79
+ - Each turn is `{"from": "...", "value": "..."}`.
80
+ - `from` is either **`human`** (the prompt) or **`gpt`** (the ground-truth response).
81
+ - The token **`<image>`** inside a human turn marks where the screenshot is spliced into the
82
+ prompt — replace it with the actual image when tokenizing.
83
+
84
+ ### Coordinate system
85
+
86
+ - Answers are **normalized to the `0–1000` range** relative to image width/height.
87
+ - A **point** answer is formatted `(x,y)`; a **bounding box** answer is `(x1,y1,x2,y2)`.
88
+ - Prompts ending in `(with point)` expect a point; `(with bbox)` expect a box.
89
+ - `unnormalized_box`, when present, is the same box in **raw pixels** — combine it with `wxh`
90
+ (or the image's true size) to convert between pixels and the normalized grid.
91
+
92
+ ## Example
93
+
94
+ ```json
95
+ {
96
+ "id": "autogui_webui_widgetlist_107976",
97
+ "conversations": [
98
+ {
99
+ "from": "human",
100
+ "value": "<image>\nPlease list all the elements on the UI screen."
101
+ },
102
+ {
103
+ "from": "gpt",
104
+ "value": "0 StaticText '3个月考上北大法硕,她说普通人成功很简单' (26,14,708,64)\n1 link '' (731,14,972,90)\n2 StaticText '时间:2022-07-20 18:11' (26,71,405,88)\n3 StaticText '老鼠吓退大猫抢吃猫粮 猫连连后退拱手相让' (26,121,692,171)\n4 link '' (731,121,972,197)\n5 StaticText '时间:2022-07-20 17:32' (26,178,405,194)\n6 StaticText '普通人镜头下的真生活' (26,227,462,251)\n7 link '' (731,227,972,303)\n8 StaticText '时间:2022-07-20 16:15' (26,284,405,301)\n9 StaticText '加载更多' (413,333,587,355)\n10 emphasis '' (26,416,33,440)\n11 StaticText '更多' (887,418,959,437)\n12 StaticText '吴亦凡打卡的面馆被整改' (26,461,505,485)\n13 link '' (731,461,972,537)\n14 StaticText '时间:2022-07-20 18:34' (26,518,405,534)\n15 StaticText '广州人才入户年龄放宽5年 本科40岁硕士45岁博士50岁' (26,568,690,617)\n16 link '' (731,568,972,643)\n17 StaticText '时间:2022-07-20 17:58' (26,624,405,641)\n18 StaticText '兵马俑手指在美被折断盗走' (26,674,549,698)\n19 link '' (731,674,972,750)\n20 StaticText '时间:2022-07-20 17:03' (26,731,405,748)\n21 StaticText '凤凰周刊:如果不靠高考,你该如何体面地出人头..' (26,781,679,831)\n22 link '' (731,781,972,857)\n23 StaticText '时间:2022-07-20 17:03' (26,838,405,854)\n24 StaticText '王中磊夫妇为儿子庆生 威廉弟弟长高了许多' (26,887,692,937)\n25 link '' (731,887,972,963)\n26 StaticText '时间:2022-07-20 16:16' (26,944,405,961)"
105
+ }
106
+ ],
107
+ "unnormalized_box": [
108
+ [
109
+ 30,
110
+ 36,
111
+ 828,
112
+ 162
113
+ ],
114
+ [
115
+ 855,
116
+ 36,
117
+ 1137,
118
+ 228
119
+ ],
120
+ [
121
+ 30,
122
+ 180,
123
+ 474,
124
+ 22
125
+ ... (truncated)
126
+ ```
127
+
128
+ ## Usage
129
+
130
+ ```python
131
+ import json, os, zipfile
132
+ from PIL import Image
133
+ from huggingface_hub import hf_hub_download
134
+
135
+ repo = "HongxinLi/UIPro-WebUI-Data-v1"
136
+ samples = json.load(open(hf_hub_download(repo, "WebUI_scale1000_488k.json", repo_type="dataset")))
137
+
138
+ images_zip = hf_hub_download(repo, "WebUI_scale1000_488k_images.zip", repo_type="dataset")
139
+ with zipfile.ZipFile(images_zip) as zf:
140
+ zf.extractall("images/")
141
+
142
+ s = samples[0]
143
+ print(s["conversations"])
144
+ img = Image.open(os.path.join("images", s["image"])) # screenshot for this sample
145
+ print("image size:", img.size)
146
+ ```
147
+
148
+ ## About UIPro
149
+
150
+ UIPro is a generalist GUI agent trained on 20.6M understanding tasks across 13 task types,
151
+ followed by agent continued fine-tuning. See the project repository for the full data
152
+ pipeline, training recipes and evaluation scripts:
153
+ **https://github.com/ZJULiHongxin/UIPro**
154
+
155
+ ## License
156
+
157
+ Released under **CC BY-NC 4.0** (non-commercial research use). The underlying screenshots and
158
+ annotations remain subject to the terms of their original source, [WebUI](https://huggingface.co/datasets/biglab/webui-all).
159
+
160
+ ## Citation
161
+
162
+ ```bibtex
163
+ @inproceedings{uipro2025,
164
+ title = {UIPro: A Generalist GUI Agent},
165
+ author = {Li, Hongxin and others},
166
+ booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
167
+ year = {2025}
168
+ }
169
+ ```