Jason Yu

@mattierialgirl Cheesed? Don't you mean you're le fromaged?

Dang cheesed I didn’t submit or publish this past year, especially with all the resources at Meta, but at least my 50K lines of code and 100 pgs of handwritten notes tell me that I tried… Hopefully I can see this ambitious diffusion project through.

How many comic sans characters did you manage to sneak past your supervisors onto your #ICCV2023 poster? I got 3.

@ZarkoValtchev We generate our results using ancestral sampling with 2000 steps. With a batch size of 1 on an RTX6000 that takes roughly 2 minutes per frame, with larger batch sizes it averages to 1 minute per frame. Methods like DDIM can further speed up things.

@JasonYujjyu What’s the inference time for a frame

Excited to present our work on Long-Term Photometric Consistent Novel View Synthesis with Diffusion Models at #ICCV2023! I hope to see you tomorrow between 14:30-16:30 at "Foyer Sud" 081!! Also checkout our project page: yorkucvil.github.io/Photoconsisten…

This work was completed at @LassondeSchool @yorkuniversity and was supported by @VectorInst, @vistayorku, and @NSERC_CRSNG. Special thanks to @MatthewKowal9 with the fastest LaTeX fingers this side of the 407. 6/6

In addition to showing that our method is competitive with previous methods on standard metrics, we use TSED to show that our method generates views that are more consistent on in-distribution, and out-of-distribution trajectories. 5/6

Happy to share our work "Long-Term Photometric Consistent Novel View Synthesis with Diffusion Models" appearing at #ICCV2023! Joint work with my AMAZING group @FereshtForghani @CSProfKGD @MarcusABrubaker!! arXiv: arxiv.org/abs/2304.10700 Project page: yorkucvil.github.io/Photoconsisten… 1/6

import torch as np

.@AirCanada after a 6 hour delay at the New Orleans airport you cancelled our flight to YYZ. At first you claimed it was delayed due to the plane coming in late & then said mechanical. Now you claim weather & are denying a hotel voucher. Does this look like bad weather to you?

Hanging out with the @yorkuniversity @LassondeSchool Computational Vision and Imaging Lab (CVIL) at #CVPR2022 #NewOrleans 💪

It's all @CSProfKGD!

This work was started as part of an internship @BorealisAI supported by @MitacsCanada and finished @LassondeSchool @yorkuniversity and @VectorInst with funding from @vistayorku and @NSERC_CRSNG. 6/6

The conditional structure of Wavelet Flow fortuitously allows it to perform super resolution. 5/6

I am excited to share our new work to appear at #NeurIPS2020! (with @CSProfKGD and @MarcusABrubaker) Wavelet Flow is a fast multi-scale normalizing flow with a conditional structure based on wavelets. Preprint: arxiv.org/abs/2010.13821 Project page: yorkucvil.github.io/Wavelet-Flow 1/6