Hi Yuji, We will not depend on Pytorch. There is a convenience function available (by the author of tensorboard-pytorch) for transforming models in onnx format to the graph protobuf in TensorFlow. If we can export MXNet symbols and params to the onnx format, we can just use that function directly without implementing the transform function from MXNet models to TensorFlow's Graph protobuf. The export function should also come out soon by other community members.
Thanks, Jun On Mon, Mar 5, 2018 at 11:56 PM, 大嶋悠司 <yuji.oshima0x...@gmail.com> wrote: > Hi Jun > > Thank you for the reply and to share your work! That's great! > I agree with using tensorboard for visualization and I made small PR for > Histogram callback( https://github.com/apache/incubator-mxnet/pull/9994 ) > Though I worried about that the dmlc/tensorboard repo would be maintained, > by looking your branch I clearly understood that you will make not to > depend on that repo. > I agree with your design and willing to contribute it in any way. > I want to add the visualization of graphs support. > According to your document, it's TBD and depends on ONNX-mxnet. > Does the reason of depending on ONNX is that you would like to use > "torch.onnx._optimize_trace" method? > > Thanks, > Yuji > > 2018-03-06 15:36 GMT+09:00 Jun Wu <wujun....@gmail.com>: > >> Hi Yuji, >> >> Welcome to the MXNet community! >> >> We have been working on delivering a complete logging solution in MXNet >> for users to employ the TensorFlow's TensorBoard for visualization. You can >> find our work in the following links: >> - Design Doc: https://cwiki.apache.org/confluence/display/MXNET/Loggi >> ng+MXNet+Data+for+Visualization+in+TensorBoard >> <https://cwiki.apache.org/confluence/display/MXNET/Logging+MXNet+Data+for+Visualization+in+TensorBoard> >> - GitHub Working Branch: https://github.com/rem >> inisce/mxnet/tree/tensorboard_logging >> >> It's mostly functional, and we are adding unit tests, tutorials, and >> documentation for APIs to make the code mergeable to the MXNet master >> branch. >> >> In order not to duplicate efforts on both sides, could you share your >> work plan so that we can decide on how to move forward? >> >> Thanks, >> Jun >> >> On Mon, Mar 5, 2018 at 10:02 PM, 大嶋悠司 <yuji.oshima0x...@gmail.com> wrote: >> >>> Hi, >>> >>> I am Yuji from NTT Research team to work on improving logging and >>> visualization of MXNet. >>> Would you help to add me into the Slack channel? Thanks. >>> >>> Best Regards, >>> Yuji Oshima >>> >> >> >
