EWSegNet: Effective Waste Segmentation

Contributions

Key Contributions

01
We propose a novel end-to-end effective waste segmentation network (EWSegNet) which improves the model's computational efficiency without compromising the segmentation performance.
02
Our frequency context module (FCM) optimally captures global context by using data-dependent kernels in the frequency domain to improve segmentation results.
03
The proposed spatial context module (SCM) performs feature excitation and weighting to complement the FCM for rich feature extraction.
04
To emphasize the boundaries of waste items and blob regions, we design the auxiliary feature enhancement module (AFEM) that uses difference of Gaussian filtering and pooled attention, thereby improving waste segmentation performance in cluttered scenes.

Overall Framework

EWSegNet Architecture

Overall framework of the proposed effective waste segmentation network (EWSegNet). The encoder consists of four stages that provide multiscale feature representations (F₁, F₂, F₃, F₄). Each stage (i) contains N_i number of EWFE layers (where i ∈ [1,2,3,4]). Before each stage, a convolution layer is used to downsample the feature maps. Feature representations of stage three are fed to auxiliary feature enhancement (AFE) module to emphasize boundaries and blob regions and obtain feature maps F₅. Finally, these multiscale features are fed to the decoder to obtain the segmentation map.

Modules

Efficient Waste Feature Extraction Layer

Efficient waste feature extraction (EWFE) layer is shown in Fig. a). Fig. b) represents the spatial context module (SCM) that is used for feature excitation and weighting in the spatial domain. In c), frequency context module (FCM) is illustrated that captures global contextual relationship between pixels in the frequency domain.

Auxiliary Feature Enhancement Module (AFEM)

This figure demonstrates the auxiliary feature enhancement module (AFEM) that has dual functions: boundaries enhancement (BE) and blob amplification (BA). BE emphasizes the fine details by using difference of Gaussian filtration while BA uses pooled attention to focus on semantic regions.

Results

State-of-the-art on ZeroWaste-f

Performance comparison of EWSegNet with state-of-the-art waste segmentation methods. Encoder FLOPs are reported with RGB image size of 512×512.

Bold — Best

Underline — Second best

EWSegNet (Ours)

Method	Encoder Params (M) ↓	GFLOPs ↓	Latency (ms) ↓	mIoU (%) ↑	Pix. Acc. (%) ↑
ReCo	—	—	—	52.28	89.33
DeepLabv3+	—	—	—	52.13	91.38
FANet	36.0	30.3	74.5	54.89	91.41
FocalNet-B	88.7	80.6	—	54.26	91.28
COSNet	27.3	24.4	73.6	56.67	91.91
EWSegNet Ours	23.3	20.5	64.8	56.44	91.75

Results

Class-wise IoU on ZeroWaste-f

Class-wise IoU (%) comparison on ZeroWaste-f dataset.

Method	Background	Cardboard	Soft Plastic	Rigid Plastic	Metal
DeepLabv3+	91.02	54.47	63.18	24.82	27.14
COSNet	91.44	59.13	65.92	37.24	29.61
EWSegNet Ours	91.45	59.24	63.17	33.28	35.05

Results

State-of-the-art on ZeroWaste-aug

mIoU (%) and mF1 (%) comparison on ZeroWaste-aug dataset.

Method	mIoU (%) ↑	mF1 (%) ↑
TopFormer-S	52.53	66.87
SeaFormer-S	54.47	68.75
AFFormer-B	54.76	69.03
PIDNet-S	57.74	71.13
FeedFormer-B0	59.18	72.58
DDRNet-slim	61.12	74.35
DeepLabv3+	52.50	—
LWCHNet	63.16	76.03
EWSegNet Ours	74.10	84.31

Results

State-of-the-art on SpectralWaste

Performance comparison in terms of mIoU (%) and class-wise IoU (%) on the SpectralWaste dataset.

Method	mIoU (%) ↑	IoU (%) ↑
Method	mIoU (%) ↑	Film	Basket	Cardboard	Video Tape	Filament	Trash Bag
MiniNet-v2	44.5	63.1	58.9	55.4	30.6	10.0	49.2
SegFormer-B0	48.4	66.9	71.3	48.9	33.6	15.2	54.6
InternImage-T	47.99	42.38	82.80	69.10	41.39	16.50	35.77
FANet	67.83	72.47	82.98	75.26	41.28	67.65	67.36
COSNet	69.96	77.61	83.65	75.14	42.95	69.06	71.38
EWSegNet Ours	71.03	77.88	84.16	79.77	42.05	73.39	68.96

Qualitative Analysis

Visual Comparison on ZeroWaste Dataset

Qualitative comparison of EWSegNet with the recent waste segmentation methods FANet and COSNet on ZeroWaste-f. Proposed EWSegNet provides reasonably better segmentation performance as highlighted in yellow boxes.

Visual Comparison on Spectral Waste Dataset

Visual comparison of EWSegNet with recent waste segmentation methods FANet and COSNet on Spectral Waste dataset. As highlighted in yellow boxes, proposed EWSegNet is fairly better to segment the waste objects in cluttered scenes.

Citation

BibTeX

@inproceedings{javaid2026ewsegnet,
  title   = {Towards Effective Waste Segmentation for Automated
             Waste Recycling in Cluttered Background},
  author  = {Javaid, Mamoona and Noman, Mubashir and Hannan, Abdul
             and Nawaz, Shah and Fiaz, Mustansar and Ghuffar, Sajid},
  booktitle = {International Conference on Machine Learning (ICML)},
  year    = {2026}
}