cannot visualize .pbtxt file

[Johnqczhang]

Hi, I'm trying to visualize a network saved in net.pbtxt file which was generated by my caffe2 code using netron of both web browser and MacOS App versions. However, both of them gave me the following error message:

Error loading model. Unsupported file content for extension '.pbtxt' in 'net.pbtxt'.

What am I supposed to do to make it work?

[lutzroeder]

The file will load if you rename it to predict_net.pbtxt. Multiple ML formats are using protobuf and protobuf design makes it difficult to detect the file format from contents. Note the file contains a lot of nodes and connections so it might take a very long time to render. A warning is shown for graphs with more than 1500 nodes, this file is close having 1460 nodes (see #68).

[Johnqczhang]

The file will load if you rename it to predict_net.pbtxt. Multiple ML formats are using protobuf and protobuf design makes it difficult to detect the file format from contents. Note the file contains a lot of nodes and connections so it might take a very long time to render. A warning is shown for graphs with more than 1500 nodes, this file is close having 1460 nodes (see #68).

Thanks a lot!

[Johnqczhang]

The file uploaded previously was generated by caffe2 when training a model with 2 GPUs so it contains nodes on all GPUs (including gradient ops), which may cause the visualization very clutter. So, I upload a new file predict_net.pbtxt.txt which contains nodes with ops for inference only on single GPU, which can give a clear visualization of the network structure for better understanding. The network was proposed in DensePose as a baseline model, which consists of a ResNet-50 as backbone network and three branches for person detection and dense pose estimation. More details can be found in the paper and its code.

BTW, both uploaded files end with .txt instead of .pbtxt because the latter one is not supported by the github.