search

cientgu / InstructDiffusion

PyTorch implementation of InstructDiffusion, a unifying and generic framework for aligning computer vision tasks with human instructions.
Other
389 stars 20 forks source link

Questions about the post-processing for keypoint detection and segmentation #9

Open essunny310 opened 1 year ago

essunny310 commented 1 year ago

Hi, very interesting work!

As mentioned in the paper, a lightweight U-Net is leveraged to post-processes the output image to generate some heatmaps (for keypoint detection) or masks (for segmentation). Which specific model did you use? How did you train the U-Net? It would be helpful if you could provide more details about this.

Thank you.

Gengzigang commented 1 year ago

Extracting keypoints or masks from the images generated by our editor is a simple mapping process, so the UNet will have minimal impact on performance.

essunny310 commented 1 year ago

Thanks for your reply.

Could you please provide more details about this procedure? Thanks a lot.