4 ms·
> The original paper used several medical X-ray datasets which I don’t have access to anymore, so I needed a new dataset with spatial annotations to test the ex
by 1970-01-01 7mo ago
>
The original paper used several medical X-ray datasets which I don’t have access to anymore, so I needed a new dataset with spatial annotations to test the expert attention mechanism. I picked the Ukiyo-eVG dataset: ~11K Japanese woodblock prints
That's such a weird switch. There's lots of free medical imaging online. Example:
https://www.cancerimagingarchive.net/ https://www.cancerimagingarchive.net/
- ykumards 7mo agoThat’s true! It felt a bit flippant to give medical data to an agent. Also, I wanted to see if the model would work in other domains!
- make3 7mo agobut doesn't it break the assumption that it should ideally be able to reproduce your original results
- ykumards 7mo agoIMO it would be hard to reproduce the results using autoresearch setup. To get CLIP to work properly we typically need large batch sizes. So the experiments in the original paper were quite heavy, and ran parallel across 8 GPUs.