| # Core deep learning | |
| torch==2.5.1 | |
| torchvision==0.16.1 | |
| # MMCV and MMDetection (video segmentation) | |
| mmcv>=2.0.0rc4,<2.1.0 | |
| mmdet==3.3.0 | |
| # Video/image processing | |
| opencv-python | |
| scipy | |
| numpy | |
| pillow | |
| # Gradio UI | |
| gradio==4.24.0 | |
| # Model/config management | |
| omegaconf==2.0.6 | |
| pyyaml | |
| # Hugging Face utilities | |
| huggingface-hub | |
| # Progress bar | |
| tqdm | |
| # fairseq (from GitHub for full compatibility) | |
| git+https://github.com/facebookresearch/[email protected] | |
| # COCO API for mmdet | |
| pycocotools | |
| # Additional utilities and metrics | |
| sacrebleu | |
| editdistance | |
| bert-score | |
| # fairscale is only needed if you use pipeline parallelism with fairseq (optional) | |
| # fairscale | |
| # For video/frame processing in some pipelines | |
| ffmpeg-python | |
| # Data manipulation and logging (if used) | |
| pandas | |
| wandb | |
| # If you use transformers models (e.g., Hugging Face GPT2 in fairseq) | |
| transformers | |
| # If you use timm or einops (common in vision models) | |
| timm | |
| einops | |
| # If you use h5py, imgaug, or fvcore (seen in some requirements in your repo) | |
| h5py | |
| imgaug | |
| fvcore | |
| # If you use spaCy (seen in requirements) | |
| spacy | |
| # If you use pyav (for advanced video I/O) | |
| pyav | |
| # If you use scikit-learn or gdown (seen in requirements) | |
| scikit-learn | |
| gdown | |
| # If you use bitarray or py3_wget (seen in requirements) | |
| bitarray | |
| py3_wget | |
| # Optional: Add any other Python packages you use below |