The Steps of StyleGaussian

Here is some of my lab training

How to train Your Own Models

1
2
3
Step 1️⃣ train_reconstruction.py    --> 生成 point_cloud.ply
Step 2️⃣ train_feature.py           --> 提取特征
Step 3️⃣ train_artistic.py          --> 风格迁移训练

1.Get point-cloud Data

This step can be finished in Orininal Gaussian Train

image.png

Reference the Catalog of Datasets/ dir

2.Reconstruct the Point Cloud

1
python train_reconstruction.py -s datasets/train

Output Model saved in: output/train/reconstruction/

3.Feature Embedding

1
2
python train_feature.py -s datasets/train \
  --ply_path output/train/reconstruction/default/point_cloud/iteration_30000/point_cloud.ply

Output Model saved in: output/train/feature

4.Style Transfer Training

Before this step, modify the train_artistic.py

and add the following code to the import part of the code:

1
2
from PIL import ImageFile
ImageFile.LOAD_TRUNCATED_IMAGES = True

or the broken image will stop the training.

then run the following command in terminal:

1
2
3
python train_artistic.py -s datasets/train \
--wikiartdir datasets/wikiart \
  --ckpt_path output/train/feature/default/chkpnt/feature.pth --style_weight 10

5.Start Viewer

1
2
python viewer.py -m output/Furiren/artistic/default \
                      --style_folder images --viewer_port 8080

view it in localhost:8080

comments powered by Disqus
Built with Hugo
Theme Stack designed by Jimmy