This is a very cool paper!
It’s also indexed on Papers with Code where you can chat with it and explore the 18 different evals on various benchmarks
Find it here:
Let's remove VAEs and ViTs from video models!
🚀 Introducing 𝗣𝗶𝘅𝗲𝗹𝗨𝗠𝗠: an encoder-free unified multimodal model for image and video understanding and generation, directly in pixel space.
Code and model available today!
🌐
📄