Gasformer
Checkpoints for Gasformer: A Transformer-based Architecture for Segmenting Methane Emissions from Livestock in Optical Gas Imaging.
Paper: arXiv | CVF | IEEE Xplore | Code: GitHub | Data: Hugging Face | Project: AI for Greener Livestock
Checkpoints
| Folder | Dataset | Checkpoint | Config |
|---|---|---|---|
CR/ |
CR (Dairy Cow Rumen Gas), binary | CR_best_mIoU_iter_160000.pth |
gasformer-160k_CR-512x512.py |
MR/ |
MR (Controlled Methane Release), 10 flow-rate classes and background | MR_best_mIoU_iter_160000.pth |
gasformer-160k_MR-512x512.py |
Each checkpoint is the best-mIoU checkpoint of a 160k-iteration MMSegmentation run at 512x512. It was trained on the official split of the dataset (official_split.csv).
Usage
git clone https://github.com/toqitahamid/Gasformer.git
hf download toqi/Gasformer --local-dir weights
python tools/test.py weights/CR/gasformer-160k_CR-512x512.py weights/CR/CR_best_mIoU_iter_160000.pth
The data are in the Gasformer dataset. Run unpack_to_png.py from the dataset repo to write the PNG folders that the code reads.
License
License: CC BY-NC 4.0. Commercial use is not permitted.
Citation
If you use these checkpoints, you must cite the paper below.
@inproceedings{sarker2024gasformer,
author = {Sarker, Toqi Tahamid and Embaby, Mohamed G. and Ahmed, Khaled R. and AbuGhazaleh, Amer},
title = {Gasformer: A Transformer-based Architecture for Segmenting Methane Emissions from Livestock in Optical Gas Imaging},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
month = {June},
year = {2024},
pages = {5489--5497},
doi = {10.1109/CVPRW63382.2024.00558}
}
Dataset used to train toqi/Gasformer
Viewer โข Updated โข 9.35k