Jinpeng Wang

hi ak: Thanks for your attention. I'd like to share this project in Hugging Face Hub. I'll apply a Space and push our code in this platform.

No Problem! I'm a liitle busy these days. I will upload the pretrain model from the server in a few days as far as possible. Besides, I plan to update...

Hi, everyone. Really sorry for late reply, the readme is update now! @jaypatravali I have provided the self-supervised pretrain model in google driver. Look Readme for details. @Mrbishuai 使用ffmpeg或者opencv 抽帧....

@NadineBehrmann try it again

Paper title: All in One: Exploring Unified Video-Language Pre-training Paper link: https://arxiv.org/abs/2203.07303 Code link: https://github.com/showlab/all-in-one Paper title: Position-guided Text Prompt for Vision Language Pre-training Paper link: https://arxiv.org/abs/2212.09737 Code link: https://github.com/sail-sg/ptp

I solved this problem by adding below in first line of nginx.conf. `user root` **It's Ridha Ben Njima's answer.** https://stackoverflow.com/questions/42388727/ffmpeg-exec-command-not-running-in-nginx-rtmp/45880931#45880931

Meet the same problem, it seems the training loss on MMC4 is hard to convergence.

Thanks. What's the sim thresh score used for this figure?

Thanks. In addition, have you ever show the validation loss before? If I use a subset of mmc4 for pretrain (around 1m website). The validation loss begin to rise up...

The downstream performance are also unstable. The ckpt in middle sometimes better than the final. I guess it's because I train the model with only 1M LAION and 1M MMc4....