Skip to content

Latest commit

 

History

History
67 lines (48 loc) · 1.69 KB

README.md

File metadata and controls

67 lines (48 loc) · 1.69 KB
Raw

If you use the code, cite the following paper:

@article{benbihi2019image,
  title={Image-Based Place Recognition on Bucolic Environment Across Seasons
    From Semantic Edge Description},
  author={Benbihi, Assia and Geist, Matthieu and Pradalier, C{\'e}dric},
  journal={Preprint},
  year={2019}
}

Run on a simple example

Coming soon.

This file describes how to run WASABI for retrievan on a CMU-Seasons example. To reproduce the paper results, please follow the instructions in the README_paper.md.

Example

The following steps describe how to retrieve images with wasabi.

They are adapted ro run on traversal 0 captured by camera 0 on slice 24 but can be easily generalised.

Image segmentation

Download the segmentation output for the example images. We use the segmentation model from https://github.com/maunzzz/cross-season-segmentation. It is a PSP-Net trained on cityscapes and finetuned on CMU-Seasons to make segmentation consistent across seasons. See their paper [1] for more details.

wget <get image segmentation>

If you want to run your own segmentation, we provide a simple script to run the segmentation model above.

git submodule init third_party/cross-season-segmentation
git submodule update

Get the weights

cd meta/weights/
./get_seg_weights.sh

Provides a path to the bgr images list and an output directory, then run:

python seg.py

Run retrieval

Edit the image directory and the segmentation directoy in scripts/wasabi_example.sh, then run

./scripts/wasabi_example.sh

Compute metrics

Display top-k retrieval

Stashed changes

Reproduce the paper results

Follow README_paper.md.