search

Tianxiaomo / pytorch-YOLOv4

PyTorch ,ONNX and TensorRT implementation of YOLOv4
Apache License 2.0
4.47k stars 1.49k forks source link

Is this really YOLOv4 ? #249

Open yuxx0218 opened 4 years ago

yuxx0218 commented 4 years ago

No CIOU Loss, no Lable Smoothing, no DropBlock Regularization, no Self-Adversarial Training, no DIOU NMS, no....... It is defnitely a YOLO v3 replacing the backbone/neck/head.

doduythao commented 4 years ago

Maybe he's still working on it

sumanthratna commented 4 years ago

I think there is support for CIOU: https://github.com/Tianxiaomo/pytorch-YOLOv4/blob/fe45fb64ba4c21130643f9b2569da7bb667356b3/train.py#L41-L127

yuxx0218 commented 4 years ago

I think there is support for CIOU:

https://github.com/Tianxiaomo/pytorch-YOLOv4/blob/fe45fb64ba4c21130643f9b2569da7bb667356b3/train.py#L41-L127

Thank you. I think you are right. Look forward to a complete version of YOLO v4.

jpc commented 3 years ago

It seems to me that bboxes_iou are used only for matching prediction and ground truth bboxes? The actual loss function seems to use only binary cross entropy and mse . Am I missing something here? https://github.com/Tianxiaomo/pytorch-YOLOv4/blob/4ccef0ec8fe984e059378813e33b3740929e0c19/train.py#L265-L267