# Create virtual environment
conda env create -f environment.yml -n grim Download TaskGrasp Dataset: Link (5GB)
gdown --fuzzy https://drive.google.com/file/d/1aZ0k43fBIZZQPPPraV-z6itpqCHuDiUU/view
unzip data.zip -d TaskGraspDownload Feature Memory: Link
git lfs install
git clone https://huggingface.co/datasets/chinchinati/GRIM_feature_mesh .
# Unzip the downloaded files
unzip TG.zip -d TG
unzip memory/fm_output.zip -d memory/fm_output# Make sure your datasets are in the same structure.
.
├── eval_grim.py # Main evaluation script
├── environment.yml # Conda environment configuration
├── memory/ # Memory storage
│ ├── memory.json # Memory data
│ └── fm_output/ # Feature memory outputs for different objects
│ ├── basket/ # Basket feature data
│ ├── fork/ # Fork feature data
│ ├── spray_nozzle/ # Spray nozzle feature data
│ ├── tiller/ # Tiller feature data
│ └── ... # etc.
├── runs/ # Experiment runs and results (each run is saved with a unique VERSION identifier)
├── TaskGrasp/ # TaskGrasp Dataset directory
│ ├── data/
│ │ ├── class_list.pkl
│ │ └── taskgrasp/scans/
├── TG/ # Processed TaskGrasp Meshes with features
├── memory/ # Memory
├── all_GT.json # Ground truth grasp idx, preprocessed from TaskGrasp for ease
└── inference.json # Inference configurations
python eval_grim.py- Evaluation Code
- Processed Memory Data
- Processed TaskGrasp Data
- Memory Creation Pipeline
- TaskGrasp Preprocessing Pipeline
- Simplified Demo
