Installation Guide
# Create and activate environment
conda create -n vision python=3.10 && conda activate vision
# Install PyTorch (CUDA or CPU version)
conda install pytorch torchvision torchaudio pytorch-cuda=11.8 -c pytorch -c nvidia -y
# or
conda install pytorch torchvision torchaudio cpuonly -c pytorch -y
# Install dependencies
pip install -r requirements.txt
# For CBIR functionality
conda install faiss-gpu=1.8.0 -c pytorch
# Optional: Install Arial font for faster inference
mkdir -p ~/.config/DuKe && cp misc/Arial.ttf ~/.config/DuKe- [Oct. 2024] Content-Based Image Retrieval(CBIR) support added with ConvNext backbone
- [Apr. 2024] Face Recognition Task(FRT) launched with various backbones and loss functions
- [Jun. 2023] Image Classification Task(ICT) released with advanced training strategies
- [May. 2023] Initial release of VisionDK
| Category | Methods |
|---|---|
| Optimization | SAM, Progressive Learning, OHEM, Focal Loss, Cosine Annealing |
| Regularization | Label Smoothing, Mixup, CutOut |
| Attention & Visualization | Attention Pool, GradCAM |
| Face Recognition | ArcFace, CircleLoss, MegFace, MV Softmax |
| Model Family | Variants |
|---|---|
| MobileNet | v2, v3_small, v3_large |
| ShuffleNet | v2_x0_5, v2_x1_0, v2_x1_5, v2_x2_0 |
| ResNet & ResNeXt | 18, 34, 50, 101, 152, 50_32x4d, 101_32x8d, 101_64x4d |
| ConvNext | tiny, small, base, large |
| EfficientNet | b0-b7, v2_s, v2_m, v2_l |
| Swin Transformer | tiny, small, base (v1 & v2) |
| Tool | Description | Usage |
|---|---|---|
| Data Splitter | Split dataset into train/val sets | python tools/data_prepare.py --postfix <jpg|png> --root <path> --frac <ratio> |
| Query-Gallery Prep | Prepare data for image retrieval | python tools/build_querygallery.py --src <path> --frac <ratio> |
| Augmentation Visualizer | Visualize data augmentations | python -m tools.test_augment |
| Data Deduplicator | Remove duplicate entries | python tools/deduplicate.py |
- For contributions: Submit a pull request
- For questions or issues: Open an issue