Aleksei Bochkovskii

Or is it correct as written here? https://blog.csdn.net/linmingan/article/details/77885832 For Focal Loss (when **gamm=2**), delta is: * `if (i == j)` then `delta = (1-p)*` **-1** `*alpha * (1 - pt)...

@jxingyizhou Hi, Greate work! https://www.zpascal.net/cvpr2019/Zhou_Bottom-Up_Object_Detection_by_Grouping_Extreme_and_Center_Points_CVPR_2019_paper.pdf > We use flip augmentation for testing. In our main comparison, we use additional 5⇥ multi-scale (0.5, 0.75, 1, 1.25, 1.5) augmentation. Finally, Soft-NMS [1]...

@Isha8 Use this repository: https://github.com/AlexeyAB/darknet And do this: https://github.com/AlexeyAB/darknet/issues/954#issuecomment-393609188

@joyivan `valid=` param isn't used during training at this moment. Use the same param `valid=` first for validation dataset, and when the optimal weights-file is found, change it to a...

@chrisrapson You use old version of Darknet. Try to update your code from GitHub: https://github.com/AlexeyAB/darknet

@chrisrapson Can you show screenshot of this error? And what command do you use?

@chrisrapson It isn't directly related to mAP calculation. * Do you use the latest version of this repo? * What parameters do you use in the Makefile? * What GPU...

@chrisrapson I would recommend you install cuDNN for CUDA 9.2 and set `CUDNN=1` in the Makefile. It may solve your issue.

@srhtyldz You can use this repository: https://github.com/AlexeyAB/darknet Just run with `-dont_show` flag: `./darknet detector demo cfg/coco.data cfg/yolo.cfg yolo.weights -dont_show test.mp4`

@BrianLiu3650 How much percent do you rate the speed of OpenCL compared to VHDL/Verilog on Intel A10 FPGA? It seems that, for example, OpenCL (SDAccel) is ~100x slower compared to...