Open Whoo-jl opened 1 year ago
Hi @Whoo-jl
There are a lot of changes need to be done to integrate new model. I will mark the issue as enhancement, the model looks interesting.
Any updates on this? There have been a lot of performance improvements in recent implementations of SAM such as EfficientSAM reducing the hardware requirements to run models for pre labelling.
I'm also interested in this enhancement. I also would be happy to do the work myself to implement but unsure what changes are needed.
Any advice would be greatly appreciated
I've implemented SAM-HQ in my own fork. It works but in my opinion it's not much better than regular SAM currently. I have yet to quantize the model which will give a boost in speed. Any suggestions for how it can be improved would be appreciated.
SAM-hq maintains SAM's original promptable design, efficiency, and zero-shot generalizability, but the image encoding module has two feature vector outputs. If I want to replace SAM with SAM-hq, what modifications do I need to make?
I added a const "interm_embeddings" in the file "/cvat/cvat-ui/plugins/sam_plugin/src/ts/index.tsx", but after compiling and restarting CVAT, the page shows an error message "Error: invalid input 'interm_embeddings'".