RCFusion

Original implementation of the paper "Recurrent Convolutional Fusion for RGB-D Object Recognition": https://arxiv.org/pdf/1806.01673.pdf

Requirements:

• Tensorflow 1.10.0
• CUDA 8.0
• Python 2.7
• See python requirements in requirements.txt
• Docker + nvidia-docker [optional]: https://github.com/NVIDIA/nvidia-docker

Instructions:

1. Download dataset and parameteres (see link below) and extract them in directory <dataset_dir> and <params_dir> [Skip to point (4) to run w/o docker]
2. To execute the code within a docker container, run docker build -t <container_name> .
3. Start the container with docker run -it --runtime=nvidia -v <dataset_dir>:<dataset_dir> -v <params_dir>:<params_dir> <container_name> bash
4. Run python code/train_and_eval.py <dataset_dir> <params_dir>

Disclaimers:

• The paper should be cosidered the main reference for this work. All the details of the algorithm and the training are reported there.
• The data augmentation taken from an external repo. Credits go to: https://github.com/aleju/imgaug
• WARNING: code has been developed w/ Tensorflow 1.5.0. We noticed some fluctuation in the results when migrating to Tensorflow 1.10.0.

Download:

• Pre-processed semantic crops of OCID dataset: https://data.acin.tuwien.ac.at/index.php/s/e46X2cCIjLXoRn7
• Original semantic crops of OCID dataset (we recommend using the link above instead to have the exact same pre-training we used in our experiments): https://www.acin.tuwien.ac.at/vision-for-robotics/software-tools/object-clutter-indoor-dataset/
• ResNet-18 pre-trained weights for initialization: https://data.acin.tuwien.ac.at/index.php/s/RueHQUbs2JtoHeJ

Contributors:

• Mohammad Reza Loghmani - email: [email protected]
• Mirco Planamente - email: [email protected]

Citation:

@ARTICLE{rcfusion, 
author={M. R. {Loghmani} and M. {Planamente} and B. {Caputo} and M. {Vincze}}, 
journal={IEEE Robotics and Automation Letters}, 
title={Recurrent Convolutional Fusion for RGB-D Object Recognition}, 
year={2019}, 
volume={4}, 
number={3}, 
pages={2878-2885}, 
doi={10.1109/LRA.2019.2921506}}