Qwen3-VL-Outpost

Qwen3-VL-Outpost is a Gradio-based web application for vision-language tasks, leveraging multiple Qwen vision-language models to process images and videos. It provides an intuitive interface for users to input queries, upload media, and generate detailed responses using advanced models like Qwen3-VL and Qwen2.5-VL.

---

Features

• Image and Video Inference: Upload images or videos and input text queries to generate detailed responses.

• Multiple Model Support: Choose from the following models:

  • Qwen3-VL-4B-Instruct
  • Qwen3-VL-8B-Instruct
  • Qwen3-VL-4B-Thinking
  • Qwen2.5-VL-3B-Instruct
  • Qwen2.5-VL-7B-Instruct

• Customizable Parameters: Adjust advanced settings such as max new tokens, temperature, top-p, top-k, and repetition penalty.

• Real-time Streaming: View model outputs as they are generated.

• Custom Theme: Uses a tailored SteelBlueTheme for an enhanced user interface.

• Example Inputs: Predefined examples for quick testing of image and video inference.

---

Installation

Prerequisites

• Python 3.10 or higher
• Git
• CUDA-compatible GPU (recommended for optimal performance)

---

Steps

1. Clone the Repository

git clone https://github.com/PRITHIVSAKTHIUR/Qwen3-VL-Outpost.git
cd Qwen3-VL-Outpost

2. Create a Virtual Environment (optional but recommended)

python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install Dependencies

Install the required packages using:

pip install -r requirements.txt

requirements.txt includes:

git+https://github.com/huggingface/accelerate.git
git+https://github.com/huggingface/peft.git
transformers-stream-generator
transformers==4.57.1
huggingface_hub
albumentations
qwen-vl-utils
pyvips-binary
sentencepiece
opencv-python
docling-core
python-docx
torchvision
supervision
matplotlib
pdf2image
num2words
reportlab
html2text
xformers
markdown
requests
pymupdf
loguru
hf_xet
spaces
pyvips
pillow
gradio
einops
httpx
click
torch
fpdf
timm
av

---

4. Run the Application

Start the Gradio interface with:

python app.py

This will launch the web interface, accessible via your browser. The application supports queuing with a maximum size of 50.

---

Usage

1. Select a Model: Choose one of the available Qwen models from the radio buttons.

2. Upload Media: Use the image or video upload section to provide input media.

3. Enter Query: Input your text query in the provided textbox.

4. Adjust Settings: Optionally tweak advanced parameters like max new tokens or temperature in the accordion.

5. Submit: Click the Submit button to generate a response.

   • Outputs are displayed in real-time in the Raw Output Stream and as formatted Markdown.

---

Example Queries

Image Inference

• “Explain the content in detail.” (with an uploaded image)
• “Jsonify Data.” (for images with tabular data)

Video Inference

• “Explain the ad in detail.” (with an uploaded video)
• “Identify the main actions in the video.”

---

Project Structure

Qwen3-VL-Outpost/
│
├── app.py              # Main application script containing the Gradio interface and model logic
├── images/             # Directory for example image files
├── videos/             # Directory for example video files
├── requirements.txt    # List of dependencies required for the project
└── README.md           # Project documentation

---

Notes

• The application uses PyTorch with GPU acceleration (torch.cuda) if available; otherwise, it falls back to CPU.
• Video processing downsamples videos to a maximum of 10 frames to optimize memory usage.
• Ensure sufficient disk space and memory when loading large models such as Qwen3-VL-8B-Instruct.
• The application is designed to run in a browser via Gradio's web interface.

---

Contributing

Contributions are welcome! To contribute:

1. Fork the repository.

2. Create a new branch:

   git checkout -b feature-branch

3. Make your changes and commit:

   git commit -m "Add new feature"

4. Push to the branch:

   git push origin feature-branch

5. Open a pull request.

---

License

This project is licensed under the Apache License 2.0. See the LICENSE file for details.