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
666 stars 62 forks source link

The paper also update original point position for over reconstructed points. #38

Closed wanmeihuali closed 1 year ago

github-actions[bot] commented 1 year ago

Running experiment on sagemaker with git sha 36091ef2de78c360a1f772da01cf2a1e39604a66

github-actions[bot] commented 1 year ago

Training job fix-densify-new-position-36091ef-230623-033516-tat-truck created

github-actions[bot] commented 1 year ago

Training job fix-densify-new-position-36091ef-230623-033516-tat-truck completed.

Model url: s3://taichi-3d-gaussian-splatting-log/tat-truck/fix-densify-new-position-36091ef-230623-033516-tat-truck/output/model.tar.gz,

tensorboard output path: s3://taichi-3d-gaussian-splatting-log/tat-truck/fix-densify-new-position-36091ef-230623-033516-tat-truck/output/output.tar.gz

github-actions[bot] commented 1 year ago

Training job fix-densify-new-position-36091ef-230623-033516-tat-truck final metrics:

Latest Metrics

train:iteration train:psnr val:loss val:ssim val:psnr train:num_valid_points train:l1loss train:loss train:ssim train:ssimloss
30000.0 25.45663070678711 0.06440120935440063 0.824465811252594 24.211393356323242 317185.0 0.030162831768393517 0.05568458512425423 0.8422284126281738 0.15777158737182617

Max Metrics

train:psnr val:7kpsnr val:psnr train:7kpsnr val:ssim val:7kssim val:5kssim train:7kssim val:5kpsnr train:5kpsnr train:5kssim train:ssim
25.45663070678711 22.7305908203125 24.211393356323242 23.7746639251709 0.824465811252594 0.785245954990387 0.7590992450714111 0.7966864109039307 21.93593978881836 23.166732788085938 0.769896924495697 0.8422284126281738