Edit2Percieve AI

⚡ Advanced Multi-Task Visual Intelligence System for Depth Estimation, Normal Map Generation, and Image Matting.

🎯 Features

• Depth Estimation: Generate high-quality depth maps from RGB images
• Normal Eeneration: Extract surface normal information from images
• Interactive Matting: Intelligent foreground/background separation
• Interactive UI: Beautiful Gradio-based web interface with real-time visualization
• CLI Support: Command-line interface for batch processing

📋 Requirements (Recommend)

• Python 3.12
• CUDA-capable GPU (recommended)
• 40GB+ VRAM for optimal performance

🚀 Installation

1. Clone the repository

git clone https://github.com/showlab/Edit2Perceive.git
cd Edit2Perceive

2. Install dependencies

pip install -r requirements.txt

3. Download Base Model

Download the FLUX.1-Kontext-dev model and place it in your desired directory:

/path/to/FLUX.1-Kontext-dev/

4. Download Our Models

Download our pre-trained models and place them in the ckpts/ directory:

ckpts/
├── edit2percieve_depth.safetensors
├── edit2percieve_normal.safetensors
└── edit2percieve_matting.safetensors

💻 Usage

Option 1: Web Interface (Recommended)

Launch the interactive Gradio UI:

python app.py

Configuration:

• Edit the model_root path in app.py (line 37) to point to your FLUX.1-Kontext model directory
• Open your browser and navigate to http://localhost:7860
• Upload an image, select a task (Depth/Normal/Matting), and click Execute

Features:

• 🎨 Interactive image editor with brush/eraser tools
• 🔍 Side-by-side comparison slider
• ⚙️ Adjustable inference parameters
• 🖱️ Point-based annotation for matting tasks

Option 2: Command Line Interface

Run inference without GUI:

python inference.py

Configuration: Edit the __main__ section in inference.py:

if __name__ == "__main__":
    # Set your model root path
    model_root = "/path/to/FLUX.1-Kontext-dev"
    
    inference(
        model_root=model_root,
        task="depth",  # Options: "depth", "normal", "matting"
        input_paths="samples/cat.jpg"  # Single image or comma-separated paths
    )

Parameters:

• model_root: Path to FLUX.1-Kontext model directory
• task: Task type - "depth", "normal", or "matting"
• input_paths: Input image path(s)
• resolution: Processing resolution (default: 768)
• num_inference_steps: Number of diffusion steps (default: 8)
• seed: Random seed for reproducibility (default: 42)
• output_path: Custom output path (optional)

📁 Project Structure

open_source_infer/
├── app.py                  # Gradio web interface
├── inference.py            # CLI inference script
├── requirements.txt        # Python dependencies
├── ckpts/                  # Model checkpoints directory
│   ├── edit2percieve_depth.safetensors
│   ├── edit2percieve_normal.safetensors
│   └── edit2percieve_matting.safetensors
├── samples/                # Sample images
├── pipelines/              # Inference pipelines
├── models/                 # Model architectures
├── trainers/               # Training utilities
└── utils/                  # Helper functions

🎨 Examples

Depth Estimation

inference(
    model_root="/path/to/FLUX.1-Kontext-dev",
    task="depth",
    input_paths="samples/cat.jpg"
)

Normal Map Generation

inference(
    model_root="/path/to/FLUX.1-Kontext-dev",
    task="normal",
    input_paths="samples/dog.jpg"
)

Image Matting

inference(
    model_root="/path/to/FLUX.1-Kontext-dev",
    task="matting",
    input_paths="samples/cat.jpg"
)

⚙️ Model Configuration

The models are configured in MODEL_CONFIGS dictionary:

MODEL_CONFIGS = {
    "Depth": {
        "path": "ckpts/edit2percieve_depth.safetensors",
        "task": "depth"
    },
    "Normal": {
        "path": "ckpts/edit2percieve_normal.safetensors",
        "task": "normal"
    },
    "Matting": {
        "path": "ckpts/edit2percieve_matting.safetensors",
        "task": "matting"
    },
}

🙏 Acknowledgments

This project is built upon the FLUX.1-Kontext model architecture.

---

Present by 🥥🍉