search

sooyekim / Deep-SR-ITM

Official repository of Deep SR-ITM (oral at ICCV 2019)
101 stars 16 forks source link

Not an issue #5

Closed Auth0rM0rgan closed 4 years ago

Auth0rM0rgan commented 4 years ago

Hey @sooyekim,

Nice work and thanks for sharing your code! I wonder if you can help me with visualizing intermediate feature maps as you did in your paper (Figure 8).

Thanks!

sooyekim commented 4 years ago

Hi @Auth0rM0rgan!

For visualizing the feature maps, you should first set net.conserveMemory to False in the testing options, to save the intermediate variables.

net.conserveMemory = False;

In the provided code, this option is set to True to save memory space. Note that depending on your GPU, there may be a memory error after changing this option. In this case, you should try cropping the input image into a smaller resolution.

With this set up, the intermediate feature maps will be saved in net.vars and you can bring it to the CPU using the 'gather' function like I did for the variable pred. For visualization, you can use imshow() with appropriate normalization for each feature map.

Best, Soo Ye