Video wird geladen...
Video konnte nicht geladen werden
๐ข๐ข๐ข ๐๐๐๐๐ฅ๐๐ซ๐๐ญ๐ข๐ง๐ ๐๐๐ฎ๐ซ๐๐ฅ ๐ ๐ข๐๐ฅ๐ ๐๐ซ๐๐ข๐ง๐ข๐ง๐ ๐ฏ๐ข๐ ๐๐จ๐๐ญ ๐๐ข๐ง๐ข๐ง๐ by Shakiba Kheradmand et al. TL;DR: importance sampling for accelerating your novel view synthesis workloads (...yes, it should also work for 3DGS)
21,810 Aufrufe โข vor 2 Jahren โขvia X (Twitter)
3 Kommentare

If you care mostly about low-PSNR workloads (think 3D perception), the ๐ ๐๐ข๐ง๐ฌ ๐๐ซ๐ ๐ฆ๐๐ฌ๐ฌ๐ข๐ฏ๐ (10x). Lead by Shakiba Kheradmand. With @DanielRebain, @beakywings, Hossam Isack, Abhishek Kar, and @kwangmoo_yi Google+SFU+UBC=โค๏ธโ๐ฅ

Expand the possibilities of your metabolic research. Resipher tracks real-time cellular oxygen consumption in standard 96-well plates, delivering continuous real-time data directly from your incubator. Request a free virtual demo or quote today >>

Cool! This is actually an under-explored area. We also tried something similar by sampling pixels with high entropy and camera rays with good baseline, instead of random. I'd say these sampling strategies are quite important with limited training budgets (in terms of iterations).
