How to visualize the network structure

[JiYuanFeng]

Hi, Is there any easy way to visualize the network? like Netscope for caffe?
I can use the net_drawer in caffe2, but found it;' so hard to read the network?

[taoari]

The following is a code snippet to visualize:

def get_model(cfg_file, weights_file):
    merge_cfg_from_file(cfg_file)
    cfg.TRAIN.WEIGHTS = '' # NOTE: do not download pretrained model weights
    cfg.TEST.WEIGHTS = weights_file
    cfg.NUM_GPUS = 1
    assert_and_infer_cfg()
    model = infer_engine.initialize_model_from_cfg()
    return model

cfg_file = '{}/configs/12_2017_baselines/e2e_mask_rcnn_R-101-FPN_2x.yaml'.format(DETECTRON_ROOT)
weights_file = 'https://s3-us-west-2.amazonaws.com/detectron/' + \
    '35861858/12_2017_baselines/e2e_mask_rcnn_R-101-FPN_2x.yaml.02_32_51.SgT4y1cO/' + \
    'output/train/coco_2014_train:coco_2014_valminusminival/generalized_rcnn/model_final.pkl'
model = get_model(cfg_file, weights_file)

from caffe2.python import net_drawer
g = net_drawer.GetPydotGraph(model, rankdir="TB")
g.write_dot(model.Proto().name + '.dot')

Then you can use a dot viewer (e.g. xdot) to visualize.

[JiYuanFeng]

Thank you taoari !
I have tested the code in my caffe2 enviorment, And I meet another problem,
File "/home/yuanfeng/Project/Detectron/lib/modeling/ResNet.py", line 95, in add_ResNet_convX_body p = model.AffineChannel(p, 'res_conv1_bn', dim=64, inplace=True) File "/home/yuanfeng/Project/Detectron/lib/modeling/detector.py", line 102, in AffineChannel return self.net.AffineChannel([blob_in, scale, bias], blob_in) File "/home/yuanfeng/Project/caffe2-master/build/caffe2/python/core.py", line 2046, in __getattr__ ",".join(workspace.C.nearby_opnames(op_type)) + ']' AttributeError: Method AffineChannel is not a registered operator. Did you mean: []

i think the original caffe2 don't register the AffinaChannel op,which is belonged to the detectron.
So should i added this op to the caffe2?

[taoari]

@JiYuanFeng You will need the imports in https://github.com/facebookresearch/Detectron/blob/master/tools/infer_simple.py

[JiYuanFeng]

@taoari Thank you for your help！
the problem has solved

[gadcam]

@taoari @JiYuanFeng Maybe you could do a pull request with your script so everyone could benefit from it? :)

[JiYuanFeng]

@gadcam i will do it ~

[liujing1995]

nice

[PacteraOliver]

@taoari

Hi, thank you for your code.

I use part of the code and make it as a tool.
I just start a pull request at #508

[drcege]

@PacteraOliver @taoari
Hi, could you really visualize the graph?

I tried GetPydotGraph and GetPydotGraphMinimal, but it hangs too long like this.

[PacteraOliver]

Yes, the whole program need to load the model (pkl file) and the config file.
The output file is a .dot file. If you open with xdot, the sample is:

The sample part of the graph is like that:

It does not take long time.