9211sun commited on
Commit
d971134
·
verified ·
1 Parent(s): fcd1f27

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +65 -19
README.md CHANGED
@@ -1,24 +1,70 @@
 
1
  configs:
2
- - config_name: auto
3
  data_files:
4
- - split: data
5
- path: "auto/data/**/*"
6
- - split: qa
7
- path: "auto/qa/*.json"
 
 
 
 
8
 
9
- - config_name: manual_train
10
- data_files:
11
- - split: data
12
- path: "manual_train/data/**/*"
13
- - split: qa
14
- path: "manual_train/qa/*.json"
15
 
16
- - config_name: manual_test
17
- data_files:
18
- - split: data
19
- path: "manual_test/data/**/*"
20
- - split: qa
21
- path: "manual_test/qa/*.json"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
22
 
23
- - config_name: navigation
24
- data_files: "navigation.json"
 
1
+ ---
2
  configs:
3
+ - config_name: default
4
  data_files:
5
+ - split: auto
6
+ path: auto/QA/*.json
7
+ - split: manual_train
8
+ path: manual_train/QA/*.json
9
+ - split: manual_test
10
+ path: manual_test/QA/*.json
11
+ license: cc-by-4.0
12
+ ---
13
 
 
 
 
 
 
 
14
 
15
+ # GUI-Xplore Dataset 🖥️🔍
16
+
17
+ GUI-Xplore is a novel dataset designed to improve **cross-application** and **cross-task generalization** for GUI agents. It introduces **exploration videos** to provide context-aware learning, enabling agents to adapt to new applications efficiently.
18
+
19
+ ## 📌 Dataset Summary
20
+ - **312 apps** across **33 categories** from **6 major domains** (*Entertainment, Productivity, Health, Shopping, Travel, News*).
21
+ - **115+ hours of exploration videos**, ensuring **comprehensive app interaction coverage**.
22
+ - **32,569 structured question-answer pairs** spanning **five hierarchical downstream tasks**.
23
+ - Designed to support **Large Vision-Language Models (LVLMs)** and **GUI-based AI agents**.
24
+
25
+ ---
26
+
27
+ ## 📂 Dataset Structure
28
+
29
+ ### 🔢 Data Splits
30
+ | Split | # Apps | # Screenshot | # Action | # Q&A Samples |
31
+ |--------------|--------|--------------|----------| --------------|
32
+ | Auto | 207 | 13,660 | 21,346 | 15,934 |
33
+ | Manual_train | 85 | 18,523 | 17,963 | 16,744 |
34
+ | Manual_test | 20 | 2,184 | 1,984 | 2,638 |
35
+ | Total | 312 | 34,367 | 41,293 | 35,316 |
36
+
37
+ ### 🏷️ Annotation Schema
38
+ Each app in the dataset includes:
39
+ 1. **Exploration Video** (`.mp4`)
40
+ 2. **Screenshot Sequence** (`.png`)
41
+ 3. **Action Sequence** (`.json`)
42
+ 4. **Screenshot-Action Relation** (`.json`)
43
+ 5. **View Hierarchy** (`.xml`)
44
+ 6. **Question-Answer Pairs** (`.json`)
45
+
46
+ ---
47
+
48
+ ## 🏗️ Tasks
49
+
50
+ GUI-Xplore defines **five downstream tasks** to evaluate GUI agents:
51
+
52
+ | Task | Description |
53
+ |---------------------------|-------------|
54
+ | **Application Overview** | Summarize app functionalities based on exploration. |
55
+ | **Page Analysis** | Interpret the purpose of a specific GUI screen. |
56
+ | **Application Usage** | Infer correct navigation sequences from homepage to task completion. |
57
+ | **Action Recall** | Identify past interactions within an app. |
58
+ | **Action Sequence Verification** | Verify logical operation order within an app. |
59
+
60
+ ---
61
+
62
+ ## 📥 How to Use
63
+
64
+ ```python
65
+ from datasets import load_dataset
66
+
67
+ dataset = load_dataset("9211sun/GUI-Xplore")
68
 
69
+ # Example data exploration
70
+ print(dataset["manual_test"][0])