MethaneMapper is a fast and accurate deep learning based solution for methane detection from airborne hyperspectral imagery. MethaneMapper introduces a spectral absorption wavelength aware transformer network and largest public dataset called Methane HotSpot dataset (MHS). This repository contains code for MethaneMapper, scripts to download and online tool to visualize dataset.
Satish Kumar*, Ivan Arevalo, A S M Iftekhar, B S Manjunath.
Official repository of our CVPR 2023 (Highlights) paper.
This repository includes:
- Source code of MethaneMapper.
- Pre-trained weights for methane plume bounding box detector and segmentation mask
- Scripts to download MHS dataset
- Online tool to visualize MHS dataset (BisQue)
- Code for custom data preparation for training/testing
- Code for mapping ground truth masks from CarbonMapper to AVIRIS-NG flightline
- Annotation generator to read-convert mask annotation into json.
The repository follows the structure of paper, making it easy to follow and use/extend the work. If this research is helpful to you, please consider citing our paper (bibtex below)
If this research is helpful to you, please consider citing our paper:
@inproceedings{kumar2023methanemapper,
title={Methanemapper: Spectral absorption aware hyperspectral transformer for methane detection},
author={Kumar, Satish and Arevalo, Ivan and Iftekhar, ASM and Manjunath, BS},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
pages={17609--17618},
year={2023}
}
- Linux or macOS with Python >= 3.7
- Pytorch >= 1.7.0
- CUDA >= 10.0
- cudNN (compatible with CUDA)
- Clone the repository
- Install dependencies
pip install -r requirements.txt
Please checkout to the BisQue
Please follow the tutorial MHS_dataset to download dataset
Follow the training tutorial Here
Please refer to CONTRIBUTING.md for contribution to the repository. Thank you!
MethaneMapper is released under the UCSB license. Please see the LICENSE file for more information.