Code for the RA-L (ICRA) 2021 paper "SeqNet: Learning Descriptors for Sequence-Based Hierarchical Place Recognition"

Last update: Dec 12, 2022

Overview

SeqNet: Learning Descriptors for Sequence-Based Hierarchical Place Recognition

[ArXiv+Supplementary] [IEEE Xplore RA-L 2021] [ICRA 2021 YouTube Video]

and

SeqNetVLAD vs PointNetVLAD: Image Sequence vs 3D Point Clouds for Day-Night Place Recognition

[ArXiv] [CVPR 2021 Workshop 3DVR]

Sequence-Based Hierarchical Visual Place Recognition.

News:

Jun 23: CVPR 2021 Workshop 3DVR paper, "SeqNetVLAD vs PointNetVLAD", now available on arXiv. Oxford dataset to be released soon.

Jun 02: SeqNet code release with the Nordland dataset.

Setup (One time)

Conda

conda create -n seqnet python=3.8 mamba -c conda-forge -y
conda activate seqnet
mamba install numpy pytorch=1.8.0 torchvision tqdm scikit-learn faiss tensorboardx h5py -c conda-forge -y

Download

Run bash download.sh to download single image NetVLAD descriptors (3.4 GB) for the Nordland-clean dataset ^[a] and corresponding model files (1.5 GB) ^[b].

Run

Train

To train sequential descriptors through SeqNet:

python main.py --mode train --pooling seqnet --dataset nordland-sw --seqL 10 --w 5 --outDims 4096 --expName "w5"

To (re-)train single descriptors through SeqNet:

python main.py --mode train --pooling seqnet --dataset nordland-sw --seqL 1 --w 1 --outDims 4096 --expName "w1"

Test

python main.py --mode test --pooling seqnet --dataset nordland-sf --seqL 5 --split test --resume ./data/runs/Jun03_15-22-44_l10_w5/

The above will reproduce results for SeqNet (S5) as per Supp. Table III on Page 10.

To obtain other results from the same table, expand this.

# Raw Single (NetVLAD) Descriptor
python main.py --mode test --pooling single --dataset nordland-sf --seqL 1 --split test

# SeqNet (S1)
python main.py --mode test --pooling seqnet --dataset nordland-sf --seqL 1 --split test --resume ./data/runs/Jun03_15-07-46_l1_w1/

# Raw + Smoothing
python main.py --mode test --pooling smooth --dataset nordland-sf --seqL 5 --split test

# Raw + Delta
python main.py --mode test --pooling delta --dataset nordland-sf --seqL 5 --split test

# Raw + SeqMatch
python main.py --mode test --pooling single+seqmatch --dataset nordland-sf --seqL 5 --split test

# SeqNet (S1) + SeqMatch
python main.py --mode test --pooling s1+seqmatch --dataset nordland-sf --seqL 5 --split test --resume ./data/runs/Jun03_15-07-46_l1_w1/

# HVPR (S5 to S1)
# Run S5 first and save its predictions by specifying `resultsPath`
python main.py --mode test --pooling seqnet --dataset nordland-sf --seqL 5 --split test --resume ./data/runs/Jun03_15-22-44_l10_w5/ --resultsPath ./data/results/
# Now run S1 + SeqMatch using results from above (the timestamp of `predictionsFile` would be different in your case)
python main.py --mode test --pooling s1+seqmatch --dataset nordland-sf --seqL 5 --split test --resume ./data/runs/Jun03_15-07-46_l1_w1/ --predictionsFile ./data/results/Jun03_16-07-36_l5_0.npz

Acknowledgement

The code in this repository is based on Nanne/pytorch-NetVlad. Thanks to Tobias Fischer for his contributions to this code during the development of our project QVPR/Patch-NetVLAD.

Citation

@article{garg2021seqnet,
  title={SeqNet: Learning Descriptors for Sequence-based Hierarchical Place Recognition},
  author={Garg, Sourav and Milford, Michael},
  journal={IEEE Robotics and Automation Letters},
  volume={6},
  number={3},
  pages={4305-4312},
  year={2021},
  publisher={IEEE},
  doi={10.1109/LRA.2021.3067633}
}

@misc{garg2021seqnetvlad,
  title={SeqNetVLAD vs PointNetVLAD: Image Sequence vs 3D Point Clouds for Day-Night Place Recognition},
  author={Garg, Sourav and Milford, Michael},
  howpublished={CVPR 2021 Workshop on 3D Vision and Robotics (3DVR)},
  month={Jun},
  year={2021},
}

Other Related Projects

Patch-NetVLAD (2021); Delta Descriptors (2020); CoarseHash (2020); seq2single (2019); LoST (2018)

[a] This is the clean version of the dataset that excludes images from the tunnels and red lights, exact image names can be obtained from here.

[b] These will automatically save to ./data/, you can modify this path in download.sh and get_datasets.py to specify your workdir.

Code for the RA-L (ICRA) 2021 paper "SeqNet: Learning Descriptors for Sequence-Based Hierarchical Place Recognition"

Related tags

Overview

SeqNet: Learning Descriptors for Sequence-Based Hierarchical Place Recognition

SeqNetVLAD vs PointNetVLAD: Image Sequence vs 3D Point Clouds for Day-Night Place Recognition

News:

Setup (One time)

Conda

Download

Run

Train

Test

Acknowledgement

Citation

Other Related Projects

Owner

Sourav Garg

Sub-tomogram-Detection - Deep learning based model for Cyro ET Sub-tomogram-Detection

BESS: Balanced Evolutionary Semi-Stacking for Disease Detection via Partially Labeled Imbalanced Tongue Data

EfficientNetV2 implementation using PyTorch

BOVText: A Large-Scale, Multidimensional Multilingual Dataset for Video Text Spotting

RaftMLP: How Much Can Be Done Without Attention and with Less Spatial Locality?

Bagua is a flexible and performant distributed training algorithm development framework.

Object detection GUI based on PaddleDetection

A user-friendly research and development tool built to standardize RL competency assessment for custom agents and environments.

A PyTorch port of the Neural 3D Mesh Renderer

pybaum provides tools to work with pytrees which is a concept burrowed from JAX.

Here is the implementation of our paper S2VC: A Framework for Any-to-Any Voice Conversion with Self-Supervised Pretrained Representations.

Finding Donors for CharityML

a morph transfer UGATIT for image translation.

Semantic Segmentation with SegFormer on Drone Dataset.

Solving reinforcement learning tasks which require language and vision

Codes to calculate solar-sensor zenith and azimuth angles directly from hyperspectral images collected by UAV. Works only for UAVs that have high resolution GNSS/IMU unit.

Semi-automated OpenVINO benchmark_app with variable parameters

The official implementation of A Unified Game-Theoretic Interpretation of Adversarial Robustness.

DenseCLIP: Language-Guided Dense Prediction with Context-Aware Prompting

A simple editor for captions in .SRT file extension