+ 
+
+
+
+
+ [CVPR' 2025] JarvisIR: Elevating Autonomous Driving Perception with Intelligent Image Restoration
+
+
+
+
+
+
+
+
+[Yunlong Lin](https://lyl1015.github.io/)1*♣, [Zixu Lin](https://github.com/)1*♣, [Haoyu Chen](https://haoyuchen.com/)2*, [Panwang Pan](https://paulpanwang.github.io/)3*, [Chenxin Li](https://chenxinli001.github.io/)6, [Sixiang Chen](https://ephemeral182.github.io/)2, [Kairun Wen](https://kairunwen.github.io/)1, [Yeying Jin](https://jinyeying.github.io/)4, [Wenbo Li](https://fenglinglwb.github.io/)5†, [Xinghao Ding](https://scholar.google.com/citations?user=k5hVBfMAAAAJ&hl=zh-CN)1†
+
+1Xiamen University, 2The Hong Kong University of Science and Technology (Guangzhou), 3Bytedance's Pico, 4Tencent, 5Huawei Noah's Ark Lab, 6The Chinese University of Hong Kong
+
+Accepted by CVPR 2025
+
+
+https://github.com/user-attachments/assets/d9094fba-e24c-403e-90cb-b3d2f6e48939
+
+
+
+
+ 
+ 
JarvisIR: Elevating Autonomous Driving Perception with Intelligent Image Restoration
+
+ Upload an image and let JarvisIR analyze its degradation and recommend the best restoration tools!
+ """)
+
+ with gr.Row():
+ with gr.Column(scale=1):
+ # Input image upload component
+ input_image = gr.Image(
+ type="filepath",
+ label="📸 Upload Your Image",
+ height=400
+ )
+
+ # Process button
+ process_btn = gr.Button(
+ "🚀 Analyze & Process",
+ variant="primary",
+ size="lg"
+ )
+
+ with gr.Column(scale=1):
+ # Chat interface to show analysis
+ chatbot = gr.Chatbot(
+ label="💬 AI Analysis Chat",
+ height=400,
+ show_label=True,
+ bubble_full_width=False
+ )
+
+ with gr.Row():
+ # Output image display
+ output_image = gr.Image(
+ label="✨ Processed Result",
+ height=300
+ )
+
+ # Connect event handler for the process button
+ process_btn.click(
+ fn=process_full_pipeline,
+ inputs=[input_image],
+ outputs=[chatbot, output_image]
+ )
+
+ # Instructions section
+ gr.Markdown("### 📝 Instructions:")
+ gr.Markdown("""
+ 1. **Upload an image** that needs restoration (blurry, dark, noisy, etc.)
+ 2. **Click 'Analyze & Process'** to let AI analyze the image
+ 3. **View the chat** to see AI's reasoning and recommendations in real-time
+ 4. **Check the result** - processed image restored to original dimensions
+ """)
+
+ return demo
+
+if __name__ == "__main__":
+ print("Starting Image Restoration Assistant...")
+ demo = create_interface()
+ # Launch the Gradio app on specified host and port
+ demo.launch(
+ server_name="0.0.0.0",
+ server_port=7866,
+ share=False
+ )
diff --git a/JarvisIR/docs/gradio_demo.md b/JarvisIR/docs/gradio_demo.md
new file mode 100644
index 0000000000000000000000000000000000000000..3d144cc9d602fe2815cbbb72c199976415c5e298
--- /dev/null
+++ b/JarvisIR/docs/gradio_demo.md
@@ -0,0 +1,36 @@
+# Gradio Demo Guide
+
+This guide provides instructions on how to run the Gradio demo for JarvisIR.
+
+## Environment Setup
+
+Please follow the environment setup instructions in the [SFT Training Guide](./sft_training.md#environment-setup) to create the conda environment and install the necessary dependencies. The same environment can be used for running the Gradio demo.
+
+Make sure you have activated the conda environment:
+```bash
+conda activate sft_jarvisir
+```
+
+
+## Download Preview Weights
+
+To run the Gradio demo, you need to download the preview weights and place them in the correct location:
+
+1. Download the JarvisIR preview weights from Hugging Face
+2. Create the weights directory (if it doesn't exist):
+ ```bash
+ mkdir -p checkpoints/jarvisir-preview/
+ ```
+3. Place the downloaded weight files in the `checkpoints/jarvisir-preview/` directory
+
+
+
+## Running the Demo
+
+Once the environment is set up and activated, you can run the Gradio demo with the following command from the root directory of the project:
+
+```bash
+python gradio_demo.py
+```
+
+This will launch a web interface where you can interact with the model.
\ No newline at end of file
diff --git a/JarvisIR/docs/sft_training.md b/JarvisIR/docs/sft_training.md
new file mode 100644
index 0000000000000000000000000000000000000000..c6c8deab0386a0b7f187b604f0575604cf0f1701
--- /dev/null
+++ b/JarvisIR/docs/sft_training.md
@@ -0,0 +1,84 @@
+# SFT (Supervised Fine-Tuning) Training Guide
+
+This guide provides step-by-step instructions for setting up the environment and performing supervised fine-tuning using XTuner framework.
+
+## Environment Setup
+
+### 1. Create and Activate Virtual Environment
+
+```bash
+conda create -n sft_jarvisir python=3.10
+conda activate sft_jarvisir
+```
+
+### 2. Install Dependencies
+
+```bash
+pip install -r requirements.txt
+cd package/agent_tools/Retinexformer && python3 setup.py develop --no_cuda_ext && cd ..
+cd RIDCP && python3 setup.py develop && cd ../..
+pip install -e .
+
+cd ../dependences/BasicSR
+pip install -e .
+
+cd ../src/sft/xtuner
+
+# Install required packages
+pip install -r requirements.txt
+
+# Install XTuner with all features
+pip install -e '.[all]'
+```
+
+### 3. Verify Installation
+
+```bash
+# Check XTuner installation
+xtuner version
+
+# Verify GPU availability
+python -c "import torch; print(f'CUDA Available: {torch.cuda.is_available()}')"
+python -c "import torch; print(f'GPU Count: {torch.cuda.device_count()}')"
+```
+
+## Model Weights Setup
+
+### 1. Create Model Directory Structure
+
+```bash
+# Create directories for models and datasets
+cd ..
+mkdir -p base_models
+```
+
+### 2. Download Base Model Weights
+
+Download the following models to the base_models folder:
+