search

tensorflow / tensorflow

An Open Source Machine Learning Framework for Everyone
https://tensorflow.org
Apache License 2.0
186.35k stars 74.31k forks source link

2.0 Reference Models: ResNet V1.5 (TPU with Keras) #25435

Closed dynamicwebpaige closed 5 years ago

dynamicwebpaige commented 5 years ago

Deep residual networks, or ResNets for short, proposed the breakthrough idea of identity mappings in order to enable training of very deep convolutional neural networks. This issue will track the creation of 1 GPU and 8 GPU TF 2.0-compatible implementations of ResNet for the ImageNet dataset.

See the following papers for more background:

[1] Deep Residual Learning for Image Recognition by Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun, Dec 2015.

[2] Identity Mappings in Deep Residual Networks by Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun, Jul 2016.

For the TensorFlow 1.x equivalent, please refer here.

sayoojbk commented 5 years ago

is someone working on this issue ?

jhseu commented 5 years ago

Yes, it'll work soon.

sayoojbk commented 5 years ago

And ya could you tell me if the implementation would be similar to the GPU version ? :https://github.com/tensorflow/tensorflow/issues/25340 Actually I wanted to make changes to the identity block of resnet like changing the dilation rate of the 3x3 conv or something like that. So will such a parameter be added or something? Cause I am actually working on a segmentation task using dilated convolution at this indentity block. If such a parameter is added it would be great if not would like to work on this task.

dynamicwebpaige commented 5 years ago

Closing out this issue, as ResNet support has landed.