Research Scientist @GoogleDeepmind - #ICCV2025 Program Chair
Cambridge MA
Joined November 2013
- Tweets76
- Following199
- Followers947
- Likes131
Deqing Sun retweeted
๐ We are excited to release the dataset, code, and expanded paper for ORBIT++: Benchmarking SfM in the Wild with 360ยฐ Video!
Following our initial publication at CVPR, weโre releasing the full benchmark and codebase:
๐ Project: orbit-sfm.github.io/
๐ arXiv: arxiv.org/abs/2608.22039
๐ป Code/Data: github.com/google-deepmind/oโฆ
(1/3)
Deqing Sun retweeted
Still struggling with frame scaling in Video LLMs? ๐คฏ
๐ซ Introducing LiteFrame: Efficient Vision Encoders Unlock Frame Scaling in Video LLMs, our work done at @GoogleDeepMind.
TL;DR: We propose LiteFrame, a highly efficient video encoder for Video Large Language Models to resolve inefficiencies in both the LLM and the ViT. [1/n]
Deqing Sun retweeted
Can dense 2D supervision help learning 4D?
Introducing UFO-4D, feedforward dense 4D reconstruction from two unposed images. By leveraging 2D signals to bridge sparse 4D datasets, we enable high-fidelity 4D interpolation in one forward pass.
๐ ufo-4d.github.io
#ICLR2026
Deqing Sun retweeted
๐ข๐ป๐ฒ ๐บ๐ฒ๐บ๐ผ๐ฟ๐ ๐ฐ๐ฎ๐ปโ๐ ๐ฟ๐๐น๐ฒ ๐๐ต๐ฒ๐บ ๐ฎ๐น๐น.
We present ๐๐ผ๐๐ฒ๐ฅ, a new ๐ต๐๐ฏ๐ฟ๐ถ๐ฑ ๐บ๐ฒ๐บ๐ผ๐ฟ๐ architecture for long-context geometric reconstruction.
LoGeR enables stable reconstruction over up to ๐ญ๐ฌ๐ธ ๐ณ๐ฟ๐ฎ๐บ๐ฒ๐ / ๐ธ๐ถ๐น๐ผ๐บ๐ฒ๐๐ฒ๐ฟ ๐๐ฐ๐ฎ๐น๐ฒ, with ๐น๐ถ๐ป๐ฒ๐ฎ๐ฟ-๐๐ถ๐บ๐ฒ ๐๐ฐ๐ฎ๐น๐ถ๐ป๐ด in sequence length, ๐ณ๐๐น๐น๐ ๐ณ๐ฒ๐ฒ๐ฑ๐ณ๐ผ๐ฟ๐๐ฎ๐ฟ๐ฑ inference, and ๐ป๐ผ ๐ฝ๐ผ๐๐-๐ผ๐ฝ๐๐ถ๐บ๐ถ๐๐ฎ๐๐ถ๐ผ๐ป.
Yet it matches or surpasses strong optimization-based pipelines. (1/5)
@GoogleDeepMind @Berkeley_AI
Deqing Sun retweeted
#ICCV2025 is deeply committed to promoting diversity, equity, and inclusion within our community. As part of this commitment, travel support is available to help broaden participation.
Applications will be reviewed on a rolling basis until August 20, 2025 (anywhere on Earth).
Deqing Sun retweeted
Ever wish you could turn your video generator into a controllable physics simulator?
We're thrilled to introduce Force Prompting! Animate any image with physical forces and get fine-grained control, without needing any physics simulator or 3D assets at inference. ๐งต(1/n)
Deqing Sun retweeted
The #ICCV2025 rebuttal deadline is almost here! Double-check the time and timezone, and be sure to submit on timeโno extensions will be granted. Good luck!
Important dates: iccv.thecvf.com/Conferences/โฆ
Deqing Sun retweeted
#KostasThoughts: When communicating with conference organizers, start by assuming good intentions and always keep it respectful. Remember, these are real people, volunteering their time and making personal sacrifices to bring everything together.
Deqing Sun retweeted
Friendly reminder: #ICCV2025 paper registration deadline is TODAY March 3rd! โฐ
Remember that paper registration is required before submission. Program chairs won't be able to accommodate late registrations.
Please check all dates and deadlines at
iccv.thecvf.com/Conferences/โฆ
A friendly reminder for those submitting to #ICCV2025 @ICCVConference: *The supplementary materials deadline is the same as the paper deadline*. (iccv.thecvf.com/Conferences/โฆ) Please make sure you are prepared.
Deqing Sun retweeted
Thanks to AI, so many amazing new possibilities are emerging.
Oh, what I would give to be a kid again. Our imagination will truly be the limit.
Deqing Sun retweeted
What happens when you train a video generation model to be conditioned on motion?
Turns out you can perform "motion prompting," just like you might prompt an LLM! Doing so enables many different capabilities. Hereโs a few examples โ check out this thread ๐งต for more results!
Deqing Sun retweeted
Code for inference, visualization, training, and evaluation is released! - GitHub.com/Junyi42/monst3r
Excited to share MonST3R! -- a simple way to estimate geometry from unposed video of dynamic scene
We achieve competitive results on several downstreams (video depth, camera pose) and believe this is a promising step toward feed-forward 4D reconstruction
monst3r-project.github.io
Deqing Sun retweeted
Do we still need codebook/quantization for scalable autoregressive visual generation?
No! Thrilled to share our latest work on scaling w/ continuous tokens. We observe power-law scaling behavior on val loss, and obtain SOTA coco FID and GenEval score.
arxiv.org/abs/2410.13863
