Title: Visual-Haptic-Kinesthetic Object Recognition with Multimodal Transformer
- python >= 3.7
- scipy
- tensorflow-gpu >= 2.5.0
- Keras >= 2.3.1
- PIL
- pandas
- Download and unzip the dataset from AU Dataset for Visuo-Haptic Object Recognition for Robots.
- Run
picenhance.py
to enhance the visual data. - Run functions in
Data_make.py
to pre-process the data. - You can request the pre-processed data from the author of this article (GitHub:Jokerr-12).
--Notes for getting started--
There is no complicated tuning of parameters for this work, and you can probably adjust the parameters of the network to achieve better results.
- run CRNN_SA method
python runtrain.py --epochs 200 --batch_size 8 --model SA
- run CRNN_CA method
python runtrain.py --epochs 200 --batch_size 8 --model CA
- test CRNN_SA save_model
python runtest.py --model SA
- test CRNN_CA save_model
python runtest.py --model CA
@InProceedings{10.1007/978-3-031-44195-0_20,
author="Zhou, Xinyuan and Lan, Shiyong and Wang, Wenwu and Li, Xinyang and Zhou, Siyuan and Yang, Hongyu",
title="Visual-Haptic-Kinesthetic Object Recognition with Multimodal Transformer",
booktitle="Artificial Neural Networks and Machine Learning -- ICANN 2023", year="2023",
publisher="Springer Nature Switzerland", address="Cham", pages="233--245", isbn="978-3-031-44195-0" }