Kanchana Ranasinghe comments

Results 12 comments of


                                            Kanchana Ranasinghe

the opl loss doesn't decrease

Hey, thanks for your interest in OPL. What is the base loss you use? Also, your base loss is not converging properly either, is it? Ideally, your loss should reduce...

Hi, thank you for the positive comments. I followed something similar to [this](https://stackoverflow.com/questions/22562364/circular-polar-histogram-in-python) to generate those polar histograms. I will try to find the exact code I used and share...

reproducing

Thank you for your interest. Also, we are extremely sorry for any lack of clarity in the repo. The backbone we use is the ViT-B/16 architecture. We simply use self-supervised...

Feature request: min/max values for logged metrics

Hey any idea if this has been added?

docker image missing requirements

Did the docker image work with those changes? I am still struggling to get this working.

get_batch() returns mmco: unref short failure, even for a valid frame

I'm facing the same issue - any updates?

Can you guys convert this loss to the tensorflow/keras version, please

Thanks a lot, we will try to add this as well.

reproducing

Sorry, we have performed self-supervised training only with Kinetics. We were hoping to explore with SSv2 as a future direction, but did not get around to it yet.

Request Demo

I am facing similar issues trying to replicate this. Sharing a demo notebook could be really useful.

student or teacher fine-tune

We use the teacher model. But our experiments showed that using the student does not lead to much difference.