Visualize the model with tensorboard

[OlwlO]

I want to visualize the model after it is built, I am trying to build the model with the build_segmentor() function and visualize the model with tensorboard.But, this approach doesn't work. How can I use this framework for model visualization?

[MeowZheng]

I'm not sure what exactly you mean by model visualization, and where did you encounter difficulties? build_segmentor or tensorboard ?

[OlwlO]

For example, after the model is built, the structure of the model can be displayed in tensorboard

[MeowZheng]

where did you encounter difficulties? build_segmentor？

[OlwlO]

Yes, I do it this way

model = build_segmentor(cfg.model)
inputs = torch.rand(1, 3, 640, 640)
with SummaryWriter(comment='model') as w:
    w.add_graph(model, (inputs, ))

However, at this point an error pops up:
TypeError: forward() missing 1 required positional argument: 'img_metas'
I'm not sure what format this img_metas is in

[MeowZheng]

I would like to suggest you use or refer to inferece_segmentor

mmsegmentation/mmseg/apis/inference.py

Line 70 in 0e37281

def inference_segmentor(model, img):

[shigejiafeimao]

I would like to suggest you use or refer to inferece_segmentor

mmsegmentation/mmseg/apis/inference.py

Line 70 in 0e37281

def inference_segmentor(model, img):

how to use? the details?

[shigejiafeimao]

I would like to suggest you use or refer to inferece_segmentor

mmsegmentation/mmseg/apis/inference.py

Line 70 in 0e37281

def inference_segmentor(model, img):

I also want to Visualize the model like the graph of tensorboard.but i dont know how to get the graph.