how can we train the "train" dataset

pierotofy / OpenSplat

Production-grade 3D gaussian splatting with CPU/GPU support for Windows, Mac and Linux 🚀

https://antimatter15.com/splat/?url=https://splat.uav4geo.com/banana.splat

GNU Affero General Public License v3.0

924 stars 87 forks source link

how can we train the "train" dataset #23

Closed james201011071 closed 8 months ago

james201011071 commented 8 months ago

hi, i test the banana dataset, but how can we test the "train" dataset, 30K iterations seems do not work well. can you give me some suggestions?

yuancaimaiyi commented 8 months ago

@pierotofy the same question ,the train data gs's result as follow: 2024-02-27 17-48-19屏幕截图 ,the num-iters set to 30000

pierotofy commented 8 months ago

The train dataset is quite noisy; if you zoom in to the train location, you should be able to see it. I used https://playcanvas.com/supersplat/editor in the past to crop out the noise before displaying it.

pierotofy commented 8 months ago

https://github.com/pierotofy/OpenSplat/issues/24 could be implemented to make it easier to locate the subjects of a scene.

pierotofy commented 8 months ago

I rendered this with 7k iterations: