search

ActiveVisionLab / DFNet

DFNet: Enhance Absolute Pose Regression with Direct Feature Matching (ECCV 2022)
https://dfnet.active.vision
MIT License
94 stars 9 forks source link

Question about how to visualize feature map in the paper? #14

Closed LZL-CS closed 10 months ago

LZL-CS commented 11 months ago

Hi @chenusc11, thanks for your awesome work. May I know how to visualize the feature map in the paper below:

image

I appreciate any help you can provide!

chenusc11 commented 10 months ago

Hi, you can save the features by normalizing feature values to (0,1) or (0,255) and save them using matplotlib

LZL-CS commented 10 months ago

Hi, you can save the features by normalizing feature values to (0,1) or (0,255) and save them using matplotlib

Got it. Do I need to use the PCA to extract the main componets, e.g., RGB 3-channels, from the feature first and then normalize it for visulization?

chenusc11 commented 10 months ago

The features are in 128 channels. You could plot one of the channels. I plotted every channel before, the contents look reasonably similar after normalization.