Tweaking to improve results

[bullmoose20]

If we see anomalies when removing backgrounds, are there tweaks we can make? I am mostly trying to remove backgrounds on people headshots and noticed that sometimes a shoulder or everything below the neck disappears...

[bullmoose20]

Before:

After (using rembg i):

After (using rembg i a -ae):

[bullmoose20]

Additionally, maybe there is a way for me to specify this somehow?
u2net_cloth_seg (download, source): A pre-trained model for Cloths Parsing from human portrait. Here clothes are parsed into 3 category: Upper body, Lower body and Full body.

And I think almost all are "Upper body"

[bullmoose20]

rembg i -m u2net_human_seg

--help does not show the -m option
README.md does not specify that there is a -m option... I stumbled on this...

[bullmoose20]

While this is a better option and I get better results, I would still like to know how I can contribute to the model to improve it. I have about 200 people headshots before and after background removal and would like to know how I can contribute to the community?

[ArielMAJ]

I ran into this same thing and u2net_human_seg helped too.

As for improving upon the model, I'm pretty sure you should be able to train your own model (or train u2net_human_seg with your own images) and use it. Pretty sure it isn't an easy task either. GL!

[alcinopitchprint]

The training process is explained here by @endh1337 , #193 (comment)