ShowUI-2B
| Property | Value |
|---|---|
| Parameter Count | 2.21B |
| Model Type | Vision-Language-Action |
| Base Model | Qwen2-VL-2B-Instruct |
| Paper | arXiv:2411.17465 |
| Tensor Type | BF16 |
What is ShowUI-2B?
ShowUI-2B is a lightweight vision-language-action model specifically designed for GUI agents. Built on the Qwen2-VL architecture, it enables sophisticated interaction with computer interfaces through visual understanding and action generation. The model represents a significant advancement in AI-driven interface manipulation, capable of understanding screen contents and executing precise actions.
Implementation Details
The model is implemented using PyTorch and utilizes Safetensors for efficient parameter storage. It features a sophisticated architecture that processes both visual and textual inputs to generate appropriate interface actions.
- Built on Qwen2-VL-2B-Instruct architecture
- Supports multiple action types including CLICK, INPUT, SELECT, HOVER, and more
- Processes images with flexible pixel requirements (256x28x28 to 1344x28x28)
- Implements coordinate-based interface interaction
Core Capabilities
- UI Grounding: Precise element location identification
- Multi-modal Navigation: Combines vision and language for interface navigation
- Action Generation: Produces contextually appropriate interface actions
- Cross-platform Support: Works with both web and mobile interfaces
- Coordinate System: Uses relative coordinates (0-1 scale) for precise positioning
Frequently Asked Questions
Q: What makes this model unique?
ShowUI-2B stands out for its specialized focus on GUI interaction, combining vision-language understanding with precise action generation in a lightweight 2B parameter package. Its ability to process screen contents and generate coordinate-based actions makes it particularly suitable for automated interface interaction.
Q: What are the recommended use cases?
The model is ideal for automated GUI testing, interface navigation assistance, and developing AI-powered interface agents. It's particularly useful for web automation, mobile app testing, and creating assistive technology for interface interaction.
