COSNet: A Novel Semantic Segmentation Network using Enhanced Boundaries in Cluttered Scenes

[Arxiv] [WACV 2025]

• Code is released !!!
• Models weights are uploaded !!!

Overview

COSNet uses boundary cues along with multi-contextual information to accurately segment the objects in cluttered scenes. COSNet introduces novel components including feature sharpening block (FSB) and boundary enhancement module (BEM) for enhancing the features and highlighting the boundary information of irregular waste objects in cluttered environment.

---

Installation

The codebase is adapted from [PVT] repository. Please follow the instructions available here to install the mmsegmentation v0.13.0.

Requirements

pytorch v1.10.1+cu111
mmsegmentation v0.13.0
mmcv 1.4.0

Training

You can find the datasets here:

[Zero-Waste-f] [Spectral-Waste]

You can utilize the below commands to train the COSNet:

Zero-Waste-f dataset

CUDA_VISIBLE_DEVICES=1 python train.py configs/cosnet/uper_cosnet_zerowaste_40k.py

Spectral-Waste dataset (RGB only)

CUDA_VISIBLE_DEVICES=1 python train.py configs/cosnet/uper_cosnet_specwaste_40k.py

ADE20k dataset

CUDA_VISIBLE_DEVICES=1 python train.py configs/cosnet/uper_cosnet_ade20k_160k.py

Evaluation

# Zero-Waste-f
python test.py configs/cosnet/uper_cosnet_zerowaste_40k.py ./zerowaste_logs/iter_40000.pth --eval mIoU

# Spectral-Waste
python test.py configs/cosnet/uper_cosnet_specwaste_40k.py ./spectralwaste_logs/iter_40000.pth --eval mIoU

# ADE20k
python test.py configs/cosnet/uper_cosnet_ade20k_160k.py ./ade20k_logs/iter_160000.pth --eval mIoU

Results

Model	Dataset	mIoU (%)
COSNet	Zero-Waste-f	56.67	download
COSNet	Spectral-Waste	69.96	download
COSNet	ADE20k	48.4	download

Model Weights are uploaded !!!

Download the pret-training ImageNet1k weights from here

Visualizations

Examples from Zero-Waste-f

Examples from ADE20k

Citation

@InProceedings{ali_2025_cosnet_WACV,
    author    = {Ali, Muhammad and Javaid, Mamoona and Noman, Mubashir and Fiaz, Mustansar and Khan, Salman},
    title     = {COSNet: A Novel Semantic Segmentation Network using Enhanced Boundaries in Cluttered Scenes},
    booktitle = {Proceedings of the Winter Conference on Applications of Computer Vision (WACV)},
    month     = {February},
    year      = {2025},
    pages     = {1363-1372}
}

@misc{ali2024cosnet,
      title={COSNet: A Novel Semantic Segmentation Network using Enhanced Boundaries in Cluttered Scenes}, 
      author={Muhammad Ali and Mamoona Javaid and Mubashir Noman and Mustansar Fiaz and Salman Khan},
      year={2024},
      eprint={2410.24139},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2410.24139}, 
}

See Also

FANet: Feature Amplification Network for Semantic Segmentation in Cluttered Background