Setup the environment with:
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
blenderproc pip install psutilNote that the cache directory for Objaverse assets has been moved in the datasets/objaverse_json.py script to avoid cluster storage limits, feel free to comment that out.
To prepare a list of Objaverse UIDs for rendering, simply run:
python datasets/objaverse_json.py --data_path <path/to/objaverse> --list_file <path/to/list.txt> --json_file <path/to/list.json> --num_objects 100 --num_workers 32The files will be downloaded and permanently cached in <path/to/objaverse>.
To be able to use ShapeNetCore, one must have access to the HuggingFace dataset. Once the dataset is saved to <path/to/shapenetcore-glb>, you can prepare a list of ShapeNet UIDs like so:
python datasets/shapenet_json.py --data_path <path/to/shapenetcore-glb> --list_file <path/to/list.txt> --json_file <path/to/list.json> --num_objects 100blenderproc download haven --categories skies high%20contrast midday --resolution 1k hdri/middayThis renderer is developed for the SLURM Workload Manager. Modify scripts/distributed.sh and simply run:
sbatch < scripts/distributed.sh