coolbrain commented on issue #6682: pre-trained resnext model doesn't work URL: https://github.com/apache/incubator-mxnet/issues/6682#issuecomment-321281288 @terrychenism I trained the resnext model according to the training log you shared. You first use 480 shorter side of image to train the model which can add more training data compared to 256, after 100 epochs, you use the 256 shorter side to improve accuracy. The top-1 of resnext-50 is 76.8%. But the top-1 of resnext-50 is 77.8% in Resnext paper implemented in Torch and the paper says they just use shorter side 256. Can Mxnet has performance as good as Torch? ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
With regards, Apache Git Services