3 ms·
I was pretty surprised to see that the WebVid-10M Dataset used as part of this training - https://m-bain.github.io/webvid-dataset/ https://m-bain.github.io/webv
by simonw 4y ago
I was pretty surprised to see that the WebVid-10M Dataset used as part of this training - https://m-bain.github.io/webvid-dataset/ https://m-bain.github.io/webvid-dataset/ - consists entirely of video preview clips scraped from Shutterstock!
I built a quick search engine over that data:
https://webvid.datasette.io/webvid/videos https://webvid.datasette.io/webvid/videos
Wrote more about that here: https://simonwillison.net/2022/Sep/29/webvid/ https://simonwillison.net/2022/Sep/29/webvid/
- thomasahle 4y agoI wonder why the generated videos don't all contain the shutterstock watermark?