Comments (2)
The current default setting actually prunes twice at iteration 16000 and 20000. (I would fix it now to align the documentation.)
There are a few arguments that you might want to play around with to get the best-performing mode with
- --prune_iterations: which is the iteration to prune the model, I would recommend having 3000 + steps after the prune for finetuning. so If you want to train a total of 12,000 steps. You could prune on 9,000 steps.
- --iteration: which is the total train iteration, you could set that to 12,000
- --densify_until_iter: This is the last iteration to densify, I recommend setting it to less than the prune iteration, else the prune would not be as effective. Maybe 8,900 if you prune at 9,000
- --prune_percent: The percent of 3d gaussians that is prune out. Since you are training fewer iterations in total you might to decrease the prune ratio.
An example command would be:
python train_densify_prune.py \
-s "PATH/TO/DATASET/$arg" \
-m "OUTPUT/PATH/${arg}" \
--prune_iterations 9000 \
--iteration 12000 \
--densify_until_iter 8900 \
--prune_percent 0.4
Play around with these settings, you could probably find a better hyperparameter :D
from lightgaussian.
I see, thank you!
from lightgaussian.
Related Issues (20)
- Getting 'tensor object is not callable' HOT 6
- new
- Experiment Settings For the FPS Measurement HOT 3
- questions about interactive reviewer HOT 3
- When will SH distillation be open sourced? HOT 1
- example ckpt for room scene HOT 1
- Error about quantizing 3DGS checkpoint HOT 1
- Errors that occur when training on the room dataset. HOT 2
- Instructions for pruning pre-trained "checkpoint" model HOT 2
- Hardware Requirements
- Regarding the size of the shared room example
- about Octree-based Compression HOT 1
- nvrtc: error: invalid value for --gpu-architecture (-arch)
- multithread add operation in kernel 'renderCUDA_count' HOT 1
- How to calculate metrics for the quantized model? HOT 2
- How to get the Global Significant Score?
- Get 1 point after excuting
- How to view?
- how to evaluate the vectree metrics (PSNR SSIM LPIPS)?
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from lightgaussian.