daniel89710

Good job! I want to try your darknet segmentation using cityscapes dataset. Could you share cfg, weight and data format for segmentation?

Special thanks.

Thank you very much. I will try it!

Thank you for your reply. I tried on Tesla V100. So, I will try on Xavier. Should I change only CUDNN_DATA_INT8x4 into CUDNN_DATA_INT8x32?

@AlexeyAB Thank you for quick update. I will check.

Hi, @AlexeyAB I tried your codes on Xavier after update code and un-comment 2 lines. However, I got a failure at checkCUDNN (transform_status) in yolo2_foward_newtork_gpu.cu :L626. Error: 3 - CUDNN_STATUS_BAD_PARAM....

@AlexeyAB Sorry for late. I tried to use -quantized flag on Xavier and I got a little speed down compared with no -quantized flag. I think we cannot still use...