Eric Ming Chen

This work was done during my internship at Snap Research, before starting my PhD, with Di Liu, @MaSizhuo60625, Michael Vasilkovsky, Bing Zhou, Qiang Gao, Wenzhou Wang, Jiahao Luo, Dimitri Metaxas, @vincesitzmann, and Jian Wang

If you're at @wacv_official tomorrow come see our poster on Snapmoji! It's our system to generate and animate 3D avatars of yourself. I'll be at poster location 51 from 4:00-5:45 on Sunday!

Check out Evan's work on scaling view synthesis transformers!

I think here it is much clearer that they are using 3dgs.

@aratahikaru0 @sichy @Olympics Yes, I think so!

Wow, it seems like the @Olympics broadcast is using Gaussian splats now. Some "spikeys" at the 10 second mark

@sichy @Olympics It's hard to see from the video compression but there are individual Gaussians visible around the skater from around 0:09 - 0:12

@ericmchen1 @Olympics Don’t think rhis is splats. I think it 1st frame last frame interpolation transfer from camera 1 to camera 2 different angles.

@RadianceFields It's really hard to tell what renders are splats and what are traditional photogrammetry techniques. But here I think the elongated Gaussian artifacts made it easy to tell

I heard splats also popped up in the ski jump, but haven’t been able to verify yet

@Jimantha Thanks Noah!

I love these time-lapses from Eric and team!

Introducing Large Video Planner (LVP-14B) — a robot foundation model that actually generalizes. LVP is built on video gen, not VLA. As my final work at @MIT, LVP has all its eval tasks proposed by third parties as a maximum stress test, but it excels!🤗 boyuan.space/large-video-pl…

Happy to finally share our latest work on Dataset Distillation! "Dataset Distillation for Pre-Trained Self-Supervised Vision Models," set to appear at #NeurIPS 2025! We learn 1 image per class to train linear heads for pre-trained models. linear-gradient-matching.github.io 1/6

@SkyLi0n I think Grain from Google is the answer. It's compatible with parquet, can load from cloud buckets, and works very nicely with data sharding

What is your favorite dataloader for large scale foundational model training? Ideally something that can process data on the fly, handle TB at scale, and has good shuffling support. Bonus points if it stores data in a generic format like parquet.

🔈@SeanXZhan and I are excited to share PhysiOpt, a differentiable physics optimizer that improves the physical integrity of 3D generative model outputs, which we will present at SIGGRAPH Asia 2025 in collaboration with Evan Thompson, Kenney Ng, and Mina Konaković Luković.

This is more apparent than ever for #ICLR2026. Reviewers accusing authors of abusing AI. Authors accusing reviewers of abusing AI. Are they right? Maybe. But regardless, everyone is looking over their shoulders and at each others' throats.

@Haian_Jin @Jimantha Congrats Haian!

So excited to share that I’ve been awarded the Google PhD Fellowship in Machine Perception! Huge thanks to my PhD advisor @Jimantha and all my amazing collaborators for their support and inspiration along the way.

Our schedule is live! curateddata.github.io/schedule.html We have posters, oral presentations, and invited talks from @liuzhuang1234, @sarameghanbeery, @andrewhowens, Phillip Isola, and Alyosha Efros! See you in Hawaii! #ICCV2025 @ICCVConference

When you think of “time-lapse video,” what usually springs to mind is a camera fixed on a tripod taking image after image at predetermined intervals. But what if you could do the same thing by taking out your phone and snapping a picture every time you walk past a certain tree on your way to work? No tripod necessary. A @CornellCIS research group has developed software that could let anyone with a camera-equipped mobile phone capture subtle changes over time – of, say, a construction site or the changing seasons – and turn them into a panoramic time-lapse video. More at news.cornell.edu/stories/2025/0….

@laurayuzheng @zzigakovacic @madhavaggar @AbeDavis Thank you!!

@ericmchen1 @zzigakovacic @madhavaggar @AbeDavis this is sooooo cool

Come see our talk on "Pocket Time-Lapse" at SIGGRAPH today at 4pm in the Image Representation, Editing, & Generation session! West Building, Rooms 118-120. With @zzigakovacic , @madhavaggar and @AbeDavis

How can we use wide-FOV cameras for reconstruction? We propose self-calibration Gaussian Splatting that jointly optimizes camera parameters, lens distortion, and 3D Gaussian representations to directly reconstruct from a set of wide-angle captures. page: denghilbert.github.io/self-cali/