hinczhang / Machine-Learning-for-3D-Geometry

Machine Learning for 3D Geometry (IN2392) from TUM.

Machine-Learning-for-3D-Geometry

please view our slides here:
Please view our report here:
It is the course of Machine Learning for 3D Geometry (IN2392) from TU Munich.

1. How to run this project?

In the SoftPool dir and the MSN dir, there are two README.md, Please read them:

• MSN is the main place to run the code, you should make and install three modules and then run the train and validation python files;
• SoftPool: Just install it and you will use this module in MSN.

2. Dataset

In the MSN/data_pre, you will find the way to handle the data. Please prepare a large space (better about 100 GB) for it.
A GPU environment is preferred as blender may accelerate via GPU.
Will not be published due to large data size. Please view ./MSN/data_pre to view the details.

3. Result

Please see the attach slides and reports.

4. Our model

Truncated SoftPool and Regional Convolution: