RobustVideoMatting VS tflite2tensorflow

Robust Video Matting/Background Remover (Remove Background from images and videos, useful for compositing) https://github.com/PeterL1n/RobustVideoMatting (RVM - Remove backgrounds from videos) https://github.com/nadermx/backgroundremover (BackgroundRemover - works well on single images) -------VOICE GENERATION--------

Looks like runway is packaging a bunch of AI tools like stable diffusion and other opensource tools into a paid package. The matting tools it is using looks like this tool https://github.com/PeterL1n/RobustVideoMatting which can be run off your computer for free if you can figure out the geeky side of installing this stuff. I've tried it out and it sometimes works well but most of the time the results aren't as good as the examples on their github. Still a good tool to have in the toolbox though.

OR this separate application looks promising, if you can work out Google Collab (I couldn't unfortunately): https://github.com/PeterL1n/BackgroundMattingV2 https://github.com/PeterL1n/RobustVideoMatting

I have done a bit of coding and I will use pretrained models only. Looking at things like depth and segmentation. Like this as an example. I am using it on a collab now but its so cumbersome. https://github.com/PeterL1n/RobustVideoMatting

I do not know much about Machine learning, and I am not sure if I can ask question here. But if yes, I need help with either choosing best libraries to do Video Editing like Background Removal and similar. Some of the ones that I found is RVM: https://github.com/PeterL1n/RobustVideoMatting (which currently seems like the best choice)

First of all I would ask if somebody retrained Robust Video Matting model on own data? I am trying to, but with all the models I end up getting bad quality result as the ones attached to the post. So my data is some objects rotating on 360 and with white backgrounds, The task seems to be pretty simple as the model just has to remove white bgr and keep colorized object. I have masks on every 10th frame of my videos. The masks are 0 - bgr, 255 - fgr. I have tried Robust Video Matting model, MODNet, PaddleSeg and several segmentation models and every of them failed to show consistent results on that data. What should I do in the case?

I have found a github with a project like this but it is tedious to use: https://github.com/PeterL1n/RobustVideoMatting

They come in the form of tflite models, so I had to convert them to onnx. I used tf2onnx for converting the pose landmark model and tflite2tensorflow for converting the pose detection model. For improving performance, I had created a small script which modified the landmark models for supporting batch inference. This script is not included in the repository, but do tell me if you need it!

What you are looking for has only been around for a month: https://github.com/PINTO0309/tflite2tensorflow