youngjae-you

@Pigdrum For example, what kind of change did you make?

@devedse When using `Cfg.use_darknet_cfg = True`, the number of filters must be modified to (5+class)*3 in case of Yolov3

When debugging, iou=0 keeps coming out of bbox loss. I think there's an issue in anchor code or between custom data and anchor box size don't match.

Candidate groups are well drawn between gt and anchor, but nothing matches (iou>0) between pred and anchor. There seems to be a problem with the model output.

@glenn-jocher Yes, what I'm curious about is whether the amount of loss calculations can vary widely because the number of predictions varies from batch to batch if the data changes...

@ebennequin If I give a temporary label to the query dataset, Isn't it predictable? But even if the metrics aren't reliable ? Can I organize dataset for interference only now?

@WongKinYiu check below issue https://github.com/WongKinYiu/PyTorch_YOLOv4/issues/390

Yolov3, v4, v5 work well

It's the same with gst-launch. Sounds like an interval problem with the nvidia deepstream Will raising the issue be a substitute to @NVIDIA? ```bash gst-launch-1.0 filesrc location = /opt/nvidia/deepstream/deepstream-6.0/samples/streams/sample_1080p_h264.mp4 !...

https://forums.developer.nvidia.com/t/pgie-interval-setting-not-work-when-use-nvdspreprocessing/200634/9?u=yeongjae8066