Harry

`loss_c = log_sum_exp(batch_conf) - batch_conf.gather(1, conf_t.view(-1, 1))` I guess this author considers `log_sum_exp` part as an approximation to maximum function, which represents predicted confidence. And `batch_conf.gather(1, conf_t.view(-1, 1))` just pick...

it seems that (at least for IForest), the calculated score is shifted with `offset`. Probably you can undo such offset for IForest to shift the score range back to 0-1,...

> Is there any reason you choose CIHP-PGN instead of Self-Correction-Human-Parsing (https://github.com/GoGoDuck912/Self-Correction-Human-Parsing) for human parsing? > > Can I use rembg (https://github.com/danielgatis/rembg) instead of carvekit to generate cloth mask? Absolutely,...

> I think they put color palette before saving the images. (https://pillow.readthedocs.io/en/stable/_modules/PIL/Image.html#Image.putpalette) Thank you for your explanation! I believe this will help others who want to reproduce this in the...

An update on the Openpose Installing section: You may want to change the last line of code if you run into some problem with cuDNN on the Colab, from ```...

> Dear lujiazho, > > Regarding human parse, the preprocessing.md says "I inferenced a parse map on 256x192 resolution, and upsample it to 1024x768. Then you can see that it...

I guess it's because you didn't convert it to P mode ``` Image.open("./output/cihp_parsing_maps/00008_00_vis.png").convert('P') ```

you may use the wrong image. You should find that the segmentation image's pixel value is not 0 - 20. For example, you need the 00568_00.png under `cihp_parsing_maps` instead of...

It seems that the demo result of "Everything" is obtained by using prompts around the whole image, which is different from using mask_generator = SamAutomaticMaskGenerator(sam) in the code.

Here is a simple implementation of web-based mask-drawer if interested: https://github.com/lujiazho/SegDrawer