search

cvat-ai / cvat

Annotate better with CVAT, the industry-leading data engine for machine learning. Used and trusted by teams at any scale, for data of any scale.
https://cvat.ai
MIT License
12.74k stars 3.03k forks source link

Support human in the loop #3329

Open SaihholdChiu opened 3 years ago

SaihholdChiu commented 3 years ago

I’m using CVAT to do an object detection labeling, my method is to label 1000 images and download to train my model , then upload predictions to CVAT and fix, do it again and again , is there any way to do all these steps in one platform?

nmanovic commented 3 years ago

@SaihholdChiu , it is something which we want to implement in the nearest future.

doantientai commented 7 months ago

Hi @nmanovic, do we have any news on this? Thank you!

doantientai commented 7 months ago

By the way, Label Studio has a workaround for this, which is integrating a model to the system so that we can train/inference it during the labeling process. Thanks to the prediction score, we sort the samples to annotate the ones with low confidence first.

Can we do something similar?

nikste commented 1 month ago

any update on this ? I am also highly interested. As far as I'm aware SAM (and SAM2) were trained with a similar way. There are some tutorials about finetuning SAM2 on custom data, that could be used for this. e.g.: https://towardsdatascience.com/train-fine-tune-segment-anything-2-sam-2-in-60-lines-of-code-928dd29a63b3

nmanovic commented 1 month ago

@nikste , @doantientai , we are working on a general approach for that. I hope to release something this year.