Use 2 man/woman LoRas

[lordsoffallen]

Hi there,

I noticed going through the code that we do segmentation based on gender mainly man and woman to differentiate the mask. I was wondering how one might go about doing 2 or more same genders using LoRas?

I saw that this showed as done in the research paper and I was curios on how this helped manage the exact position of each LoRa character? Since we do object detection, we get mask location of all man and all woman but how do align each mask position with the exact LoRa character?

[peacej]

I have the same question, seems like an important drawback of OMG. Perhaps one solution is to just let the end user click on the image to determine which character is which, and expand the mask automatically from the clicked pixel. similar to the UI of https://segment-anything.com/demo# .