boisbb / multi_view_transformer

MVT: Multi-view Vision Transformer for 3D Object Recognition

Overview

This an open-source implementation of a method for 3D model attention based classification proposed by Chen et al.

How to install dependencies

To install them, run:

pip install -r requirements.txt

How to train the model

First, specify the training configuration and all model/DataModule hyperparameters in their respective .yaml config files under configs/, then run in the root directory chose between mvt_module and mvt_module_with_back_bone in the target name of the model/mvt.yaml config file to train either model

python train.py

How to test the model

After specifying the path to the checkpointed model and its config file (respective target name for example) run in root directory:

python test.py

Authors

Project implemented and maintained by Alexandre Lutt, Boris Burkalo, Mohamed Said Derbel, Ludwig Gräf.