phd student @PrincetonVL. formerly: bse @princetoncs, research intern @elorianai

Joined March 2025
some things never change! from HN thread on CycleGAN in 2017.
2
21
2,882
great prescience from dario here
2
3
34
1,112
goodbye yellow-theme wandb 😢
3
210
good charts are parsimonious
straight lines on graphs. you can’t even see where ai happened!
1
300
found a guy on ebay who sells career fair quant merch and loose barebells
9
3
3
470
55,284
fable should be kinder to the subagents
9
467
i aspire to become a sota larper
19
1,859
Noticing a trend where zoomer founders advertise their high lol/valorant ranks. Alas, my only notable gaming accomplishment is hitting masters elo in 2018 town of salem.
8
506
Replying to @itsandrewgao
Hasn't shown up in degree data; this is based off what current sophomores are declaring. dailyprincetonian.com/articl…
5
647
I’d previously thought that single-view reconstruction would be tough with only synthetic data, but it turns out it’s not! Check out this very cool work applying procedural 3D data to *full* reconstruction.
Releasing RecGen: a collaboration between @ToyotaResearch, @toyota_europe, and @UvA_Amsterdam tackling a core 3D vision challenge: reconstructing complete multi-object scenes (parts, poses, textures, even occluded geometry) from just 1 to a few RGB-D views. Trained purely on synthetic data, RecGen achieves SOTA on real-world robotics and 6D pose benchmarks, handling occlusions, symmetry, and complex interactions. A step toward scalable, high-fidelity digital twins for robotics, and better evaluation and training of generalist policies. reconstruction-by-generation…
1
14
2,220
By collecting the best design choices from our study, we create a full-scale dataset, WMGStereo-150k. Our data is super sample efficient and scales well! [6/7]
1
5
932
We also find that training on transparent or reflective surfaces harms performance on diffuse regions, suggesting that we might need to co-design our model priors with the training data! [5/7]
1
2
664
More realistic data isn’t always better! Instead, the best data comes from a combination of realistic background scenes with added flying objects. We hypothesize that realistic scenes alone are not sufficiently diverse, while flying objects alone have large domain gap. [4/7]
1
1
1
6
757
We use Infinigen, a procedural data generator, to create a ton of small datasets with different variations to figure out what design choices really matter. We study object placements, lighting, camera parameters, and more.[3/7]
1
1
8
837
Designing a new stereo dataset is challenging. If one dataset is better than another, was it because of the scene realism, object diversity, materials, or camera placement? Existing datasets change multiple factors at the same time! [2/7]
1
7
944