Saving the out of the layer to view them as an image

1

Hi, I am working with residual networks. I want to view features of intermediated layer as an image. How do I do that if

               def residual_unit:
                return  res1, outAfterMask, mask

when I call that in my model:

                 self.res1 = resudal_unit(128, 128)

How do I save mask and view as an image. What I think if I can save them as an numpy array in .hf5 file and then view it using matplotlib. I am not sure how to save using a numpy array. Is it when I call the method in my model ? Any suggestions will be helpful

2

Hi,
You can do it with matplotlib :

plt.imshow(MyTensor.permute(1, 2, 0).cpu().detach().numpy())

Or with PIL :

from torchvision import transforms
transforms.ToPILImage()(MyTensor).show()

In any case, I think your tensor needs to be 3xWidthxHeight

1 Like
3

@lerezell Thank you for your answer. My input is a video. So, I want to see one of its frame. But my question is when I call that method, do I just call it as

                self.res1, self. outAfterMask, mask = residal_unit(128, 128)

Later,

                              output=model(input)

How do I get that MyTensor?

4

MyTensor should be any tensor (3xHxW) that you want to see.

From a video you can get it with openCv or other tools

For you architecture, I don’t know what you’re trying to do.