An plug in and play pipeline that utilizes segment anything to segment datasets with rich detail for downstream fine-tuning on vision models like CLIP, ViT, Imagebind, and so on!
2 Pipelines
2 potential pipelines, what do you think? We should make one for robotic datasets
1. Infinigen -> Segment Anything Video -> Very Rich and Detailed Dataset.
2. Segment Anything for Image and or Video -> Iterate over Dataset and segment-> Very Rich and Detailed Dataset structured dataset for pretraining?
I support 2, segment anything (SAM) from facebook was show a power in image tasks, now fine-tune and acceleration of SAM are two ways for future, so creating more and more good quantity dataset is fine work based on fine-tune SAM with hand labeling work, so Pipelines of labeling work in future will be based on cycle of fine-tune SAM workflow, handing label work change to fix the edge of labels created from SAM
aiyou9
commented
1 year ago 
I support 2, segment anything (SAM) from facebook was show a power in image tasks, now fine-tune and acceleration of SAM are two ways for future, so creating more and more good quantity dataset is fine work based on fine-tune SAM with hand labeling work, so Pipelines of labeling work in future will be based on cycle of fine-tune SAM workflow, handing label work change to fix the edge of labels created from SAM