Scholar
Vighnesh Birodkar
Google Scholar ID: whm4LjQAAAAJ
Google
Citations & Impact
Citations
3,634
H-index
12
i10-index
13
Publications
20
Co-authors
0
Publications
Resume (English only)
Academic Achievements
- - Trained auto encoders with diffusion loss (without GANs) and achieved both better compression and higher quality generation.
- - Contributed to Google's flagship AI model, Gemini 2.5 and Veo.
- - Trained an LLM with discrete video tokens for zero-shot video generation.
- - Fixed a critical flaw in Region Proposal networks to achieve strong mask generation across classes.
- - Showed that 10% of ImageNet can be discarded without any loss in evaluation accuracy.
Research Experience
- Works in the video pre-training team at Luma AI; contributed to the Gemini 2.5 and Veo projects by adding video data and metrics to the Gemini codebase and training a prototype diffusion transformer that became Veo 1.
Background
- A machine learning researcher in the space of video generative models. Deeply interested in compression, real-time, and long context generation. Has been generating videos since 2017. Currently works in the video pre-training team at Luma AI.
Miscellany
- My wife (left) and I (right) at the Mendocino arch.
Co-authors