tjr16/bathy_nn_learning

Data-driven Loop Closure Detection in Bathymetric Point Clouds for Underwater SLAM

bathy_nn_learning

Introduction

This repository is intended for loop closure detection and feature matching in the context of Multibeam Echo Sounders (MBES).

Dependencies

• Python 3.8
• PyTorch 1.10
• PyTorch Geometric 2.0
• opencv, wandb, shapely, open3d, AUVLib (original repo)

Install the June 2022 version of AUVLib here:

git clone -b extended_bm [email protected]:ignaciotb/auvlib.git

For details, please refer to requirements.txt (for pip) or environment.yml (for conda).

Recommended for baseline: PCL 1.10 or its python binding

Data

Datasets are available for download here datasets

Usage

Step 1: Run scripts/parse_cereal.py to parse cereal data.

Step 2.1 - 2.3: Run other scripts in scripts/ to create datasets.

Step 3: Run train.py to train a model. (Modify param.py properly.)

Step 4: Run scripts in test/ to evaluate the model.

.
├── data               # datasets
│   ├── Circle100      # training set
│   │   ├── raw        # raw training set
│   │   └── processed  # processed training set
│   ├── Circle100Valid
│   │   └── ...
│   └── Circle100Test
│       └── ...
├── scripts     # scripts for data processing
├── utils       # utility functions 
├── test        # testing scripts
├── models.py   # model implementation
├── dataset.py  # dataset implementation
├── param.py    # parameters and configurations
└── train.py    # training script

Citation

If you find our work useful, please consider citing:

@inproceedings{tan2023data,
    title={Data-driven loop closure detection in bathymetric point clouds for underwater {SLAM}},
    author={Tan, Jiarui and Torroba, Ignacio and Xie, Yiping and Folkesson, John},
    booktitle={2023 IEEE International Conference on Robotics and Automation (ICRA)},
    pages={3131--3137},
    year={2023},
    organization={IEEE}

}

Acknowledgment

Part of the code is based on some examples in PyTorch Geometric.

bathy_nn_learning's Issues

undefined symbol: mbes _ ping

Hello, thank you very much for this job.
When I run the generate_training_set.py file to generate a training set, I get an error: Traceback (most recent call last):
File "bathy_nn_learning/scripts/generate_training_set.py", line 85, in
processed_bin = random_sample(raw_bin, num_sample1)
File "bathy_nn_learning/utils/dataset.py", line 58, in random_sample
pcd_np_sample = pcd_np[np.random.choice(pcd_np.shape[0], size=int(n_points), replace=False), :]
File "mtrand.pyx", line 984, in numpy.random.mtrand.RandomState.choice
ValueError: Cannot take a larger sample than population when 'replace=False' I thought the size of random sampling was too large, so I changed the parameter num_sample1 = 19000 to a smaller size, but after re-running the code, num of points: min will also be smaller, which is still the above error.
Why is this? Looking forward to your answer, thank you very much.

firstly! thank you for your work!

is the dataset availble to acquire ? if no, sorry to bother you.

raise AttributeError

Hello, when I run the "test_place_recognition.py" program for the first time, I want to generate the "negative_pairs.pkl" file.
However, when calling the function "sample_negative_pairs" in the file "utils/dataset.py", the code is "_, _, train _ test _ pose, _ = pair _ dataset. Get _ train _ test (train _ idx, test _ idx)".
The prompt is as follows: the object "pair_dataset" has no attribute "get_train_test", which leads to "raise AttributeError" error.
But why click on "pair_dataset" will show that it has two attributes: "processed_dir" and "get_train_test".
Why is this happening? I would appreciate it if someone could answer my questions.