Alexey

@WongKinYiu Hi, Why did you add `CenterNet resnet-102` instead of `CenterNet dla-34` to your comparison table?

@WongKinYiu > However, I think the order of multiple-inputs-route-layer in https://github.com/AlexeyAB/darknet and https://github.com/ultralytics/yolov3 may be different. This conversion code works well with yolov3-spp.weights/cfg file despite the fact that yolov3-spp uses...

> But now training on whole dataset with Multi GPU parameters causing system reboot, This is a hardware issue: power insufficient or hardware bug in GPU.

try to train by using 2-3 GPUs instead of 4.

@atangfan The .weights-files are identical in https://github.com/AlexeyAB/darknet and https://github.com/WongKinYiu/PyTorch_YOLOv4 but **different cfg-files:** * for Darknet you should use `activation=logistic` * for PyTorch_YOLOv4 you should use `activation=linear`

You can calculate new anchors by using this command: `./darknet detector calc_anchors coco.data -num_of_clusters 9 -width 512 -height 512 -show` 

* `img = cv::Scalar::all(255);` https://github.com/AlexeyAB/darknet/blob/bef28445e57cd560fa3d0a24af98a562d289135b/src/image_opencv.cpp#L1472 * `cv::rectangle(img, pt1, pt2, CV_RGB(0, 0, 0), 1, 8, 0);` https://github.com/AlexeyAB/darknet/blob/bef28445e57cd560fa3d0a24af98a562d289135b/src/image_opencv.cpp#L1490

What is the black spare? There is no problem.

So +-0.7 AP can be just fluctuation on early stages of training. Or partial-residual connections require more iterations for training than common-residual connections?

@WongKinYiu Hi, * It seems that the (Mosaic/Smooth/Mish) affects the model CSPResNeXt-50 better than the models CSPResNet-50 and CSPDarknet-53 for Classifier 256x256, because **CSPResNeXt-50 has more outputs of each layer**...