Datasets:
Update README.md
Browse files
README.md
CHANGED
|
@@ -46,6 +46,8 @@ A key strength of this benchmark is its meticulous annotation: all bounding boxe
|
|
| 46 |
|
| 47 |
|
| 48 |
|
|
|
|
|
|
|
| 49 |
## Dataset Structure
|
| 50 |
|
| 51 |
The dataset contains 1,639 samples divided into three key groups:
|
|
@@ -72,26 +74,44 @@ The dataset includes several challenging scenarios:
|
|
| 72 |
Pixel-Navigator focuses on realism by capturing authentic interactions: actions taken by humans and agents.
|
| 73 |
The records of Pixel Navigator are English-language, desktop-size screenshots of 100+ websites. Each record points to an element outlined by a rectangular bounding box and an intent corresponding to it. In particular, the dataset focuses on providing bounding boxes and intents that are not ambiguous, thus increasing the trustworthiness of the evaluation of a VLM on this data.
|
| 74 |
|
|
|
|
|
|
|
|
|
|
| 75 |
|
| 76 |
-
|
| 77 |
|
| 78 |
With this new benchmark, H Company aims to unlock new capabilities in VLMs, and stimulate the progress of web agents.
|
| 79 |
|
| 80 |
-
###
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 81 |
|
| 82 |
-
| Category | Instruction | Image |
|
| 83 |
-
|---------|----------|-------|
|
| 84 |
-
| UI Understanding | Access user account settings | <img src="./examples/Access%20user%20account%20settings.png" width="200"/> |
|
| 85 |
-
| UI Understanding | Add Insignia cable to cart | <img src="./examples/Add%20Insignia%20cable%20to%20cart.png" width="200"/> |
|
| 86 |
-
| UI Understanding | Pick the first available date | <img src="./examples/Pick%20the%20first%20available%20date.png" width="200"/> |
|
| 87 |
-
| Format understanding | Choose 18:45 | <img src="./examples/Choose%2018_45.png" width="200"/> |
|
| 88 |
-
| UI composition | Green Button to create a travel alert | <img src="./examples/Green%20Button%20to%20create%20a%20travel%20alert.png" width="200"/> |
|
| 89 |
-
| UI composition | log in button (middle of the page) | <img src="./examples/log%20in%20button%20(middle%20of%20the%20page).png" width="200"/> |
|
| 90 |
-
| UI composition | Select fifth image in gallery | <img src="./examples/Select%20fifth%20image%20in%20gallery.png" width="200"/> |
|
| 91 |
-
| Calendar understanding | Select aug 7th | <img src="./examples/Select%20aug%207th.png" width="200"/> |
|
| 92 |
|
| 93 |
|
| 94 |
-
#### UI Understanding
|
| 95 |
|
| 96 |
|
| 97 |
|
|
@@ -131,3 +151,5 @@ All labels were hand-written or hand-reviewed. Instructions were rewritten when
|
|
| 131 |
|
| 132 |
research@hcompany.ai
|
| 133 |
|
|
|
|
|
|
|
|
|
| 46 |
|
| 47 |
|
| 48 |
|
| 49 |
+
|
| 50 |
+
|
| 51 |
## Dataset Structure
|
| 52 |
|
| 53 |
The dataset contains 1,639 samples divided into three key groups:
|
|
|
|
| 74 |
Pixel-Navigator focuses on realism by capturing authentic interactions: actions taken by humans and agents.
|
| 75 |
The records of Pixel Navigator are English-language, desktop-size screenshots of 100+ websites. Each record points to an element outlined by a rectangular bounding box and an intent corresponding to it. In particular, the dataset focuses on providing bounding boxes and intents that are not ambiguous, thus increasing the trustworthiness of the evaluation of a VLM on this data.
|
| 76 |
|
| 77 |
+
### Challenging Examples for UI Element Selection
|
| 78 |
+
|
| 79 |
+
[comment]: # (Link to presentation with images https://docs.google.com/presentation/d/1NQGq75Ao_r-4GF8WCyK0BRPCdvkjzxIE2xP9ttV5UcM/edit#slide=id.g358e1dac3df_0_60)
|
| 80 |
|
| 81 |
+
Our dataset includes examples that go beyond standard object detection or OCR, requiring genuine **UI understanding** and **instruction-based visual reasoning**. These examples highlight failure points in current models and test capabilities critical for real-world interaction with user interfaces, demonstrating H Company's commitment to creating targeted benchmarks around challenging areas.
|
| 82 |
|
| 83 |
With this new benchmark, H Company aims to unlock new capabilities in VLMs, and stimulate the progress of web agents.
|
| 84 |
|
| 85 |
+
### Key Challenges Captured in the Benchmark
|
| 86 |
+
|
| 87 |
+
- **UI Understanding**
|
| 88 |
+
Tasks require comprehension of common UI conventions (e.g., icons, labels, layout). For instance, identifying the correct user settings button may involve recognizing a gear icon, or adding a specific product to a cart might require interpreting both imagery and adjacent labels. State-of-the-art models often fail at such tasks due to lack of contextual or semantic UI awareness.
|
| 89 |
+
|
| 90 |
+
- **Instruction-Based Disambiguation**
|
| 91 |
+
Some instructions describe objects based on spatial position, appearance, or intent (e.g., "middle of the page", "green button"). These tasks demand models to combine textual instruction with visual reasoning—something most models do not yet handle robustly.
|
| 92 |
+
|
| 93 |
+
- **Calendar Navigation**
|
| 94 |
+
Even frontier models struggle to interact with calendar widgets. Understanding which dates are available (e.g., not grayed out or marked unavailable) is a frequent failure case, demonstrating gaps in dynamic UI interpretation.
|
| 95 |
+
|
| 96 |
+
- **Format and Locale Sensitivity**
|
| 97 |
+
Instructions that rely on regional formats—like time (“18:45”) or date representations—test the model’s resilience to locale-specific variations. Models trained on culturally homogeneous data often perform poorly here.
|
| 98 |
+
|
| 99 |
+
### Example Tasks
|
| 100 |
+
|
| 101 |
+
| **Category** | **Instruction** | **Image** |
|
| 102 |
+
|------------------------|------------------------------------------------|-----------|
|
| 103 |
+
| UI Understanding | Access user account settings |  |
|
| 104 |
+
| UI Understanding | Add Insignia cable to cart |  |
|
| 105 |
+
| UI Understanding | Pick the first available date |  |
|
| 106 |
+
| Format Understanding | Choose 18:45 |  |
|
| 107 |
+
| UI Disambiguation | Green button to create a travel alert |  |
|
| 108 |
+
| UI Disambiguation | Log in button (middle of the page) | .png) |
|
| 109 |
+
| UI Disambiguation | Select fifth image in gallery |  |
|
| 110 |
+
| Calendar Understanding | Select Aug 7th |  |
|
| 111 |
+
|
| 112 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 113 |
|
| 114 |
|
|
|
|
| 115 |
|
| 116 |
|
| 117 |
|
|
|
|
| 151 |
|
| 152 |
research@hcompany.ai
|
| 153 |
|
| 154 |
+
|
| 155 |
+
|