- Change model definition from functional to class
- Dynamic creation of models depending on the files in models/ (Currently you have to specify a concrete list of defined models)
- Refactoring all utils files and adding more explanation
- Change training mode from keras' fit to custom tensorflow training loop
- Add Logging
- Add wandb logging support
- Change trained model filename to one including the time such that models are not overwritten
- Add support for COCO
- Add support for KITTI
- Add support for ILSVRC DET 2017
- Add support for ILSVRC VID 2017
- add ensemble support
- add temporal ensemble support
- add multi-GPU training support
- write documentation
- Add support for YOLOv1
- Add support for YOLOv2
- Add support for YOLOv3
- Add support for YOLOv4
- MIT License
lugen4ro/object_detection_tf2
README
Contributors
