search

floe / backscrub

Virtual Video Device for Background Replacement with Deep Semantic Segmentation
Apache License 2.0
734 stars 85 forks source link

A "wider" model? #165

Open RicardoPQ-com opened 12 months ago

RicardoPQ-com commented 12 months ago

Hi there. Does anyone have a model similar to "selfiesegmentation_mlkit-256x256-2021_01_19-v1215.f16.tflite", but that doesn't crop the sides of the image? I've tried "segm_lite_v509.tflite" and "segm_lite_v681.tflite", but they don't seem to have the same resolution as the first one.

floe commented 12 months ago

Hmmm, if I remember correctly, @BenBE had some code that would just run the model twice for wider images, but I'm not sure if that ever made it to the main branch...

BenBE commented 12 months ago

Nope. That code was some early WIP and I couldn't really get the necessary refactoring to work for processing multiple images per frame.