12 comments

[ 2.9 ms ] story [ 55.0 ms ] thread
failed pretty bad on a group picture (15 people)
It was primarily trained on portrait shots. So not many samples with groups of people. May change in the future.
Gave decent results for me.

I'm guessing it cost quite a bit of time+money to build this?

Time, primarily
failed terribly on 5 different images gave up
What images did you use? Could you give me an example so I can look into this?
Tried this on ten different portraits, worked perfectly on all of them.
All my images failed. They were not portraits.
Giving good results as such. It's struggling with hair, unsurprisingly. I noted that it removes all my hair below the ears where people typically do not have it, so may be under-represented in the training data.
Great tool. This should be upvoted more. Tuxa, I'd love to discuss more about this tool with you. I'll shoot you an email when I'm free later.
Cool! But how do you expect to earn money?
This actually worked extremely well on a few images - with only some minor cleanup being needed for an image to be "usable".

Here's an example: https://vgy.me/vyqGV7.png

There's a bit of black smudge between the two women and a bit of the hair of the doctor on the bottom left was removed. A tiny portion of the woman's face on the right was removed but would be easy enough to fix. A tiny bit of the doctor's shoulder was removed but that's understandable given the contrast between it and the wall.

Saving 90% of the time and manually doing 10% of the cleanup is bloody amazing.