building a multitask training platform based on Pytorch
cls-dev branch for classfication or re-identification tasks.
- Classfication (Single-label, Multi-label) and Re-identification tasks.
- Metric learning for Re-identification.
- Visualize training curve (wandb / tensorboard).
- Remove similarity or noise data, weighted k-nearest neighbor for reid.
- Visualize Precision-Recall / Receiver-Operating-Characteristic curve.
- Mixed-Precision Training for faster speed.
- Visualize models' heatmaps and U-MAPs.
- Convert pth model to onnx / rknn / ncnn format.
1.👀 Use the State-of-the-Art image classfication toolkit.
- Various backbones and pretrained models
- Bag of training tricks
- Large-scale training configs
- High efficiency and extensibility
- Powerful toolkits
2.🚀 Enhance codes' reusability.
3.🛠️ Minimize our project.
For detailed installation guides, please refer to INSTALL.md.
- Train & Validate with Training Curve
Refer to README-Train/Val Models for details.
- Convert Pth Model To ONNX/RKNN
Refer to README-Convert Models for details.
This project is released under the Apache 2.0 license.
Feel free to create a pull request if you want to contribute (e.g. networks or tricks).