search

wgcban / adamae

[CVPR'23] AdaMAE: Adaptive Masking for Efficient Spatiotemporal Learning with Masked Autoencoders
https://www.wgcban.com/research/adamae
MIT License
72 stars 8 forks source link

How to obtain the reconstructed image for inference and masked #3

Closed hzxie99 closed 10 months ago

hzxie99 commented 10 months ago

Hello, I very much agree with your work. I would like to know how to obtain the schematic diagram of the reconstructed image and the mask image during the experiment. Because I'm just getting started. I really appreciate it if you can help me with this question.

wgcban commented 10 months ago

@hzxie99 Thanks for your interest!.

I wrote a python notebook (which I cannot locate now, but similar to one used in image MAEs https://colab.research.google.com/github/facebookresearch/mae/blob/main/demo/mae_visualize.ipynb) which

  1. first saves the predicted image and mask into numpy array
  2. then visualize each frame in the video by re-arranging and re-normalizing and then saving it as png file

I will try to find my script and post it here. But, I would say it is a modification on top of the above python notebook file.

hzxie99 commented 10 months ago

Thank you very much. The pictures in the paper are very beautiful. Thank you!!

hzxie99 commented 10 months ago

@hzxie99 Thanks for your interest!.

I wrote a python notebook (which I cannot locate now, but similar to one used in image MAEs https://colab.research.google.com/github/facebookresearch/mae/blob/main/demo/mae_visualize.ipynb) which

  1. first saves the predicted image and mask into numpy array
  2. then visualize each frame in the video by re-arranging and re-normalizing and then saving it as png file

I will try to find my script and post it here. But, I would say it is a modification on top of the above python notebook file.

I'm sorry to bother you, but I want to ask if you have found your notes and could you share an example? My ability to draw 3d mask reconstruction images is limited