search

spacepxl / ComfyUI-Florence-2

Florence-2 image captioning and tasks
https://huggingface.co/microsoft/Florence-2-large
MIT License
70 stars 6 forks source link

Can you add the functionality of combining image batches? This way, we can achieve video background removal #4

Open wibur0620 opened 5 months ago

wibur0620 commented 5 months ago

Can you add the functionality of combining image batches? This way, we can achieve video background removal image

dnl13 commented 5 months ago

It's not easy with video when using multiple masks. I've done this before in a fork for segment_anything, where the masking was quite accurate, but it involves adding multiple masks (sub-batches) to the initial batch. Here, we need to think about the most efficient approach, as the post-processors also need to be adjusted accordingly to handle potentially empty masks and, most importantly, whether the model maintains mask quality across the batch length. Wouldn't Rembg / Image Remove Background Node in ComfUI Essentials be more suitable for removing the background in your task?? It can handle image batches effectively. Just my thoughts.