Mehdi S. M. Sajjadi
@msajjadi.com
Research Scientist
Tech Lead & Manager
Google DeepMind
msajjadi.com
Tech Lead & Manager
Google DeepMind
msajjadi.com
Scaling 4D Representations
Self-supervised learning from video does scale! In our latest work, we scaled masked auto-encoding models to 22B params, boosting performance on pose estimation, tracking & more.
Paper: arxiv.org/abs/2412.15212
Code & models: github.com/google-deepmind/representations4d
Self-supervised learning from video does scale! In our latest work, we scaled masked auto-encoding models to 22B params, boosting performance on pose estimation, tracking & more.
Paper: arxiv.org/abs/2412.15212
Code & models: github.com/google-deepmind/representations4d
July 10, 2025 at 11:52 AM
Scaling 4D Representations
Self-supervised learning from video does scale! In our latest work, we scaled masked auto-encoding models to 22B params, boosting performance on pose estimation, tracking & more.
Paper: arxiv.org/abs/2412.15212
Code & models: github.com/google-deepmind/representations4d
Self-supervised learning from video does scale! In our latest work, we scaled masked auto-encoding models to 22B params, boosting performance on pose estimation, tracking & more.
Paper: arxiv.org/abs/2412.15212
Code & models: github.com/google-deepmind/representations4d
Generative Video Diffusion: does a model trained with this objective learn better features compared to image generation?
We investigated this question and more in our latest work, please check it out!
*From Image to Video: An Empirical Study of Diffusion Representations*
arxiv.org/abs/2502.07001
We investigated this question and more in our latest work, please check it out!
*From Image to Video: An Empirical Study of Diffusion Representations*
arxiv.org/abs/2502.07001
February 13, 2025 at 4:11 PM
Generative Video Diffusion: does a model trained with this objective learn better features compared to image generation?
We investigated this question and more in our latest work, please check it out!
*From Image to Video: An Empirical Study of Diffusion Representations*
arxiv.org/abs/2502.07001
We investigated this question and more in our latest work, please check it out!
*From Image to Video: An Empirical Study of Diffusion Representations*
arxiv.org/abs/2502.07001
TRecViT: A Recurrent Video Transformer
arxiv.org/abs/2412.14294
Causal, 3× fewer parameters, 12× less memory, 5× higher FLOPs than (non-causal) ViViT, matching / outperforming on Kinetics & SSv2 action recognition.
Code and checkpoints out soon.
arxiv.org/abs/2412.14294
Causal, 3× fewer parameters, 12× less memory, 5× higher FLOPs than (non-causal) ViViT, matching / outperforming on Kinetics & SSv2 action recognition.
Code and checkpoints out soon.
January 10, 2025 at 3:44 PM
TRecViT: A Recurrent Video Transformer
arxiv.org/abs/2412.14294
Causal, 3× fewer parameters, 12× less memory, 5× higher FLOPs than (non-causal) ViViT, matching / outperforming on Kinetics & SSv2 action recognition.
Code and checkpoints out soon.
arxiv.org/abs/2412.14294
Causal, 3× fewer parameters, 12× less memory, 5× higher FLOPs than (non-causal) ViViT, matching / outperforming on Kinetics & SSv2 action recognition.
Code and checkpoints out soon.