search

Tianxiaomo / pytorch-YOLOv4

PyTorch ,ONNX and TensorRT implementation of YOLOv4
Apache License 2.0
4.46k stars 1.49k forks source link

Move to YOLOV7 or YOLOX project unmaintained #557

Open ghsanti opened 1 year ago

ghsanti commented 1 year ago

-deleted-

PrajwalCogniac commented 1 year ago

Hey @mahnehsilla Please tell me this I am really struggling

ghsanti commented 1 year ago

@PrajwalCogniac use yolov7 or yolox this project is unmaintained and does not work

PrajwalCogniac commented 1 year ago

Hey @mahnehsilla Thanks a lot for the reply, So the problem I am facing is I have to use yolov4 for inference. I don't have control over the training portion. My client gave me the .cfg and weights of yolo4 and I have to run inference on it. Also the format of .cfg for route layer is like this [route] layers = -1 groups = 2 group_id = 1 Can I use the yolo7 or yolox to load the yolo4 weights and config ? Also any help in navigating this will be deeply appreciated

ghsanti commented 1 year ago

No, you cant transfer the weights to yolox or yolov7 because the Neural Net Architectures are different shape, and the weights can not be transferred to completely different shapes. The weights can only be transferred if one nn is extension of another, or the exact same shape.

It may be possible to do inference with this project, but you will have to share a notebook for me to help, and I can not be sure I can help.

The way to go is to pick yolov7 or yolox, train the NN so you get the new weights, and do inference with the model you trained i.e yolov7 or yolox.