facebookresearch / Detectron

FAIR's research platform for object detection research, implementing popular algorithms like Mask R-CNN and RetinaNet.
Apache License 2.0
26.27k stars 5.46k forks source link

How can I get the visualization of feature maps after trained with Detectron? #851

Closed CPFelix closed 5 years ago

CPFelix commented 5 years ago

I have sucessifully trained the mask r-cnn,and I want to visualize the feature maps in FPN and Res50,but I have no ideas.Did anyone have done this?thanks!

ir413 commented 5 years ago

Hi @CPFelix, the easiest may be to (1) save intermediate features for images you care about to disk (e.g. in pkl format; see #123 for how to extract feature blobs) and (2) write a tool or notebook that visualizes the pre-computed features.

zunairaR commented 3 years ago

how i can do this in pytorch. using the latest API