GasTwinFormer
Checkpoint for GasTwinFormer: A Hybrid Vision Transformer for Livestock Methane Emission Segmentation and Dietary Classification in Optical Gas Imaging (ICCV 2025 Workshops).
- Paper: https://openaccess.thecvf.com/content/ICCV2025W/SEA/html/Sarker_GasTwinFormer_A_Hybrid_Vision_Transformer_for_Livestock_Methane_Emission_Segmentation_ICCVW_2025_paper.html
- IEEE Xplore: https://ieeexplore.ieee.org/document/11374624
- Code: https://github.com/toqitahamid/GasTwinFormer
- Data: https://huggingface.co/datasets/toqi/beef-cattle-methane-ogi
- Project: https://ai-greener-livestock.github.io/
Files
| File | Description |
|---|---|
gastwinformer_best_mIoU_iter_80000.pth |
Best-mIoU checkpoint from the 80k-iteration run (weights only, no optimizer state) |
Model
- Mix Twin encoder (
MixTwinVisionTransformerV3) with a global + local attention pattern in all four stages, 32 / 64 / 160 / 256 channels, local window size 5. - Hierarchical LR-ASPP decoder with 128 channels.
- Gaussian plume-weighted Dice loss for segmentation, with a classification head for the diet label.
- 3.35 M parameters, 3.43 GFLOPs, 114.9 FPS, as reported in the paper.
Usage
Clone the code repository and load this checkpoint with its config, gastwinformer/configs/gastwinformer_80k.py:
from mmseg.apis import init_model
model = init_model("gastwinformer/configs/gastwinformer_80k.py", "gastwinformer_best_mIoU_iter_80000.pth", device="cuda:0")
License
License: CC BY-NC 4.0. Commercial use is not permitted.
Citation
If you use this checkpoint, you must cite the paper below.
@inproceedings{sarker2025gastwinformer,
author = {Sarker, Toqi Tahamid and Embaby, Mohamed and Islam, Taminul and AbuGhazaleh, Amer and Ahmed, Khaled},
title = {GasTwinFormer: A Hybrid Vision Transformer for Livestock Methane Emission Segmentation and Dietary Classification in Optical Gas Imaging},
booktitle = {2025 IEEE/CVF International Conference on Computer Vision Workshops (ICCVW)},
year = {2025},
pages = {2808--2817},
doi = {10.1109/ICCVW69036.2025.00293}
}
Acknowledgments
This work is supported by the project "Artificial Intelligence for Greener Livestock: Educational and Research", award no. 2022-70001-37404, from the U.S. Department of Agriculture's National Institute of Food and Agriculture.
Dataset used to train toqi/GasTwinFormer
Viewer โข Updated โข 11.7k โข 22