-
Filter classes to visualize during inference using the
--classes
command line argument with space separated class indices from the dataset YAML file.For example, to visualize only persons in COCO dataset, use,
python inference.py --classes 1 <rest of the command>
To visualize person and car, use,
python inference.py --classes 1 3 <rest of the command>
-
Added Deep SORT Real-Time tracking to
inference_video.py
andonnx_video_inference.py
. Using--track
command with the usual inference command. Support for MobileNet Re-ID for now.
- New DenseNet backbones.
- Mosaic augmentation updated to be Ultralytics/YOLOv5/YOLOv8 like.
- Updated augmentations regime for better training results
- Released a Mini Darknet Nano Head model pretrained on the Pascal VOC model for 600 epochs. Find the release details here.
- Can load COCO/Pascal VOC pretrained weights for transfer learning/fine tuning using the
--weights
flag and providing the path to the weights file. - Resume training by providing the path to the previous trained weights using the
--weights
flag and--resume-training
flag to continue from previous plots and load the optimizer state dictionary as well. Here,--weights
will take the path to thelast_model.pth
and notbest_model.pth
orlast_model_state.pth
. The latter two models store the model weights dictionary only. - Weights and Biases logging possible now.
- Default training resolution now 640x640. Provides slightly better results.