search

wanmeihuali / taichi_3d_gaussian_splatting

An unofficial implementation of paper 3D Gaussian Splatting for Real-Time Radiance Field Rendering by taichi lang.
Apache License 2.0
659 stars 62 forks source link

Try different tile size #128

Open wanmeihuali opened 1 year ago

wanmeihuali commented 1 year ago

the article(https://tellusim.com/07_hello_splatting/) claims that removing shared memory and inference 8 pixels per thread have a good acceleration on inference. May be we can try that. Let's first try adjust the tile size.

github-actions[bot] commented 1 year ago

Running experiment on sagemaker with git sha df61017060b31dc94da96a5e8d843ca263223dd5

github-actions[bot] commented 1 year ago

Training job try-8-pixels-perthread-df61017-230921-075944-tat-truck-baseline created

github-actions[bot] commented 1 year ago

Running experiment on sagemaker with git sha df61017060b31dc94da96a5e8d843ca263223dd5

github-actions[bot] commented 1 year ago

Training job try-8-pixels-perthread-df61017-230921-075954-tat-train-baseline created

github-actions[bot] commented 1 year ago

Training job try-8-pixels-perthread-df61017-230921-075944-tat-truck-baseline failed

github-actions[bot] commented 1 year ago

Training job try-8-pixels-perthread-df61017-230921-075944-tat-truck-baseline final metrics:

github-actions[bot] commented 1 year ago

Training job try-8-pixels-perthread-df61017-230921-075954-tat-train-baseline failed

github-actions[bot] commented 1 year ago

Training job try-8-pixels-perthread-df61017-230921-075954-tat-train-baseline final metrics: