qingzhouzhen commented on issue #7957: add densenet URL: https://github.com/apache/incubator-mxnet/pull/7957#issuecomment-331686274 The training of 169-layers is done, result as below: INFO:root:Epoch[124] Batch [2450] Speed: 107.01 samples/sec accuracy=0.904687 cross-entropy=0.365293 top_k_accuracy_5=0.984219 INFO:root:Epoch[124] Batch [2300] Speed: 106.73 samples/sec accuracy=0.902656 cross-entropy=0.367414 top_k_accuracy_5=0.985313 INFO:root:Epoch[124] Batch [2100] Speed: 106.68 samples/sec accuracy=0.897656 cross-entropy=0.385241 top_k_accuracy_5=0.983594 INFO:root:Epoch[124] Batch [2500] Speed: 106.51 samples/sec accuracy=0.903438 cross-entropy=0.372258 top_k_accuracy_5=0.984219 INFO:root:Epoch[124] Train-accuracy=0.906250 INFO:root:Epoch[124] Train-cross-entropy=0.400677 INFO:root:Epoch[124] Train-top_k_accuracy_5=0.980469 INFO:root:Epoch[124] Time cost=3106.063 INFO:root:Saved checkpoint to "densenet-models/densenet-0125.params" INFO:root:Epoch[124] Validation-accuracy=0.741744 INFO:root:Epoch[124] Validation-cross-entropy=1.160414 INFO:root:Epoch[124] Validation-top_k_accuracy_5=0.911830 I think the model is consistent with the gluon implementation, with little different such as BatchNorm layer, but it seems did not affect the result. @piiswrong ---------------------------------------------------------------- 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