It is trained using commit 0a49638 from V9-m as initial weights, obtaining a mAP in red. In the pull request #175 , it is observed that when using YOLO detection format annotation, it is considered as if it were annotated in segmentation format. Therefore, a pull request is manually implemented to handle this case, and the model is retrained, achieving a better mAP, although it still leaves much to be desired.
Reading comments from other issues, HenryTsui mentions that after 10 epochs, the model does not converge like other YOLOs.
On the other hand, the stipulated dataset format is used as a functional dataset format. #166 (comment) as a format of dataset. This mAP are usual in other types of training.
It is trained using commit 0a49638 from V9-m as initial weights, obtaining a mAP in red. In the pull request #175 , it is observed that when using YOLO detection format annotation, it is considered as if it were annotated in segmentation format. Therefore, a pull request is manually implemented to handle this case, and the model is retrained, achieving a better mAP, although it still leaves much to be desired.
Reading comments from other issues, HenryTsui mentions that after 10 epochs, the model does not converge like other YOLOs.
On the other hand, the stipulated dataset format is used as a functional dataset format. #166 (comment) as a format of dataset. This mAP are usual in other types of training.