pengzhao-intel commented on issue #8302: Refactor operators & MKLDNN URL: https://github.com/apache/incubator-mxnet/pull/8302#issuecomment-359674818 @eric-haibin-lin Yes, I think the new MKL-DNN is OK for both inference and training. The full size of ImageNet training is under the testing, it trends to coverage from the current curve. Small dataset training results as below: Dataset | Model/Network | Test script | Training accuracy | Comments -- | -- | -- | -- | -- MNIST | MLP | example/image-classification/train_mnist.py | 99.96% ? CiFAR-10 | ResNet-50 | example/image-classification/train_cifar10.py | 86.02% (26 epochs) | official rec files from MXNET sampled ImageNet | Inception-v3 | example/image-classification/train_cifar10.py | 89.20% (15 epochs) | sampled 100,200 images of 200 categories from full imagenet, and converting to rec file via img2rec.py provided by official MxNet.
---------------------------------------------------------------- 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
