search

eriklindernoren / PyTorch-YOLOv3

Minimal PyTorch implementation of YOLOv3
GNU General Public License v3.0
7.31k stars 2.64k forks source link

Does it have support on the Scaled-yolov4 model? #765

Closed kadirnar closed 2 years ago

kadirnar commented 2 years ago

parser.add_argument("-m", "--model", type=str, default="cfg/yolov4-csp.cfg") parser.add_argument("-w", "--weights", type=str, default="weights/yolov4-csp.weights")

Output-1:

teaser

Output-2:

teaser

Output-3:

teaser

How can i solve this problem?

Flova commented 2 years ago

Unitl now I have not tried Scaled-YOLOv4. Could you outline the changes regarding the computation grapth, pre and post-processing?

kadirnar commented 2 years ago

I did not understand your question. I changed the file paths in the detection file. It works on yolov3 and yolov4 models, but not on scaled-yolov4 and yolov4-mish models.

Scaled Yolov4: https://github.com/AlexeyAB/darknet/issues/7087 Orijinal Repo Scale-Yolov4: https://github.com/WongKinYiu/ScaledYOLOv4 Yolov4-Mish: https://github.com/AlexeyAB/darknet/issues/6987

kadirnar commented 2 years ago

Pytorch-Yolov3: https://github.com/eriklindernoren/PyTorch-YOLOv3/blob/master/pytorchyolo/detect.py#L19 Yolov4-Csp: https://github.com/AlexeyAB/darknet/blob/master/cfg/yolov4-csp.cfg#L1276 The nms function is used in the PyTorch-YOLOv3 repo. Used diounms in the scaled-yolov4 repo. Could it be because of this?