Mind-the-Glitch

The official implementation for the Neurips 2025 paper "Mind-the-Glitch: Visual Correspondence for Detecting Inconsistencies in Subject-Driven Image Generation"

🚀 Release Status

• Model Release - Pre-trained MTG model weights and inference code
• Training Code - Training scripts and configuration files
• Training Dataset - Automated dataset generation pipeline and curated dataset
• Evaluation Benchmark - Benchmark evaluation code and metrics

📦 Installation

Prerequisites

• Python 3.8+
• CUDA-compatible GPU (recommended)

Setup Environment

1. Clone the repository with submodules:

git clone --recursive https://github.com/abdo-eldesokey/mind-the-glitch.git
cd mind-the-glitch

2. Create and activate a conda environment:

conda create -n mtg python=3.11
conda activate mtg

3. Install PyTorch with CUDA support:

pip install torch==2.5.1 torchvision==0.20.1 --index-url https://download.pytorch.org/whl/cu124

4. Install remaining dependencies:

pip install -r requirements.txt

5. Initialize submodules (if not cloned recursively):

git submodule update --init --recursive

6. Install Grounded-Segment-Anything:

git clone https://github.com/IDEA-Research/Grounded-Segment-Anything.git
cd Grounded-Segment-Anything
pip install --no-build-isolation -e GroundingDINO
cd ..

🎯 Getting Started

The easiest way to get started with Mind-the-Glitch is through our interactive playground notebook:

jupyter notebook notebooks/playground.ipynb

This notebook demonstrates:

• Loading the pre-trained MTG model
• Running inference on sample images
• Visualizing the disentangled features and visual correspondence.

📚 Citation

If you find this work useful for your research, please cite our paper:

@inproceedings{eldesokey2025mindtheglitch,
  title={Mind-the-Glitch: Visual Correspondence for Detecting Inconsistencies in Subject-Driven Generation},
  author={Eldesokey, Abdelrahman and Cvejic, Aleksandar and Ghanem, Bernard and Wonka, Peter},
  booktitle={Advances in Neural Information Processing Systems},
  year={2025}
}