MViTac: Multimodal Visual-Tactile Representation Learning through Self-Supervised Contrastive Pre-Training

The repository was jointly developed with Vedant Dave

Project Website

This repository contains the implementation for training and evaluating the MVitAC model. The model is trained using contrastive learning on the Calandra dataset. The code was developed and tested on Ubuntu 22.04 with Python 3.10.

Overview

1. mvitac.ipynb: This notebook contains the procedures for pretraining the MViTac model using contrastive learning. It covers configuration setup, model architecture, training, and evaluation.
2. evaluate.ipynb: A notebook dedicated to evaluating pretrained models. It provides metrics and visual results to understand the model's performance.
3. contrastive_learning_dataset.py: A script that prepares datasets specifically for contrastive learning.
4. generate_dataset.py: This script generates datasets, likely preprocessing and structuring data in a way suitable for training and evaluation.
5. model.py: Defines the architecture of the MVitAC model.
6. utils.py: Contains utility functions that assist in training, evaluation, and other tasks.
7. config.py: Contains configuration settings and parameters that dictate various aspects of training and evaluation.

How to Use

1. Setup: Ensure all dependencies are installed. This can typically be done using a requirements.txt file.
2. Data Preparation: - Download the dataset from this link and place it in the mvitac/calandradataset/ folder. Use the generate_dataset.py script to prepare your data. This will preprocess and structure your data into a suitable format.
3. Training: Open the mvitac.ipynb notebook and follow the steps to pretrain the MViTac model.
4. Evaluation: Once the model is trained, use the evaluate.ipynb notebook to train a linear classifier on the top of the learned representations to evaluate its performance on test data.
5. Utilities: The utils.py script contains helper functions. It's integrated into the training and evaluation notebooks, so there's no need to run it separately.

Notes

• Ensure the data paths in the configuration are correctly set.
• Monitor the training progress to ensure convergence. Adjust hyperparameters in config.py if necessary.
• For custom datasets, ensure they are structured correctly and update data paths in the configuration.

Cite

@misc{dave2024multimodal,
  title={Multimodal Visual-Tactile Representation Learning through Self-Supervised Contrastive Pre-Training}, 
  author={Vedant Dave and Fotios Lygerakis and Elmar Rueckert},
  year={2024},
  eprint={2401.12024},
  archivePrefix={arXiv},
  primaryClass={cs.RO}
}

License

MIT License

Copyright (c) [2023] [Chair of Cyber-Physical Systems]

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.