Context enhanced stereo transformer(CSTR)

This is the official repo for our work Context enhanced stereo transformer(CSTR)

Dependencies

We recommend the following steps to set up your environment

• Create your python virtual environment by

  conda create --name CEST python=3.8 # create a virtual environment called "CEST" with python version 3.8

• Install Pytorch 1.9 and cuda 11.6:

  • Other versions of Pytorch may have problem during evaluation.

• Other third-party packages: You can use pip to install the dependencies by

  pip install -r requirements.txt

Pre-trained Models

You can download the pretrained model from the following links:

Models	Link
CSTR (Scene Flow pretrained,plugging context enhanced path to sttr-light)	Download link

• Note: the pretrained model is only trained on FlyingThings3D sub-dataset.

Folder Structure

Code Structure

Context enhanced stereo-transformer
    |_ dataset (dataloder)
    |_ module (network modules, including loss)
    |_ utilities (training, evaluation, inference, logger etc.)

Usage

• Download pretrained model using links in Pre-trained Models.

• Run pretraining by

  sh scripts/pretrain.sh
  

• Run fine-tune on dataset by

  sh scripts/finetune.sh
  

• Run evaluation on the middlebury by

  sh scripts/middlebury_eval.sh
  

  • Note: the pretrained model is assumed to be in the stereo-transformer folder.

License

This project is under the Apache 2.0 license.

Contributing

We try out best to make our work easy to transfer. If you see any issues, feel free to fork the repo and start a pull request.