cs.CVMar 28, 2026

Evaluating Large and Lightweight Vision Models for Irregular Component Segmentation in E-Waste Disassembly

Authors: Xinyao Zhang, Chang Liu, Xiao Liang, Minghui Zheng, Sara Behdad

Organizations: Florida State University, Industrial and Manufacturing Engineering, Tallahassee, FL, USA · Texas A&M University, Mechanical Engineering, College Station, TX, USA · University of Florida, Environmental Engineering Sciences, Gainesville, FL, USA

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Precise segmentation of irregular and densely arranged components is essential for robotic disassembly and material recovery in electronic waste (e-waste) recycling. This study evaluates the impact of model architecture and scale on segmentation performance by comparing SAM2, a transformer-based vision model, with the lightweight YOLOv8 network. Both models were trained and tested on a newly collected dataset of 1,456 annotated RGB images of laptop components including logic boards, heat sinks, and fans, captured under varying illumination and orientation conditions. Data augmentation techniques, such as random rotation, flipping, and cropping, were applied to improve model robustness. YOLOv8 achieved higher segmentation accuracy (mAP50 = 98.8%, mAP50-95 = 85%) and stronger boundary precision than SAM2 (mAP50 = 8.4%). SAM2 demonstrated flexibility in representing diverse object structures but often produced overlapping masks and inconsistent contours. These findings show that large pre-trained models require task-specific optimization for industrial applications. The resulting dataset and benchmarking framework provide a foundation for developing scalable vision algorithms for robotic e-waste disassembly and circular manufacturing systems.

Figures & tables

Figure 1: SAM2 architecture.
Figure 2: Yolov8 architecture
AspectSAM2YOLOv8
Design objectiveGeneral-purpose and class-agnostic segmentationTask-specific detection and segmentation
Training strategyPre-trained on SA-V dataset, and fine-tuned on target dataTrained from scratch on target dataset
Mask generationMultiple candidate masks per imageOne mask per detected object
Class assignmentRequires post-processing or promptingDirectly predicts class labels
Irregular shapesUses learned shape priors from diverse pre-trainingLeans domain-specific features
Inference speedHigher computational costLightweight and real-time capable
Table 1: Comparison of SAM2 and YOLOv8 for e-waste segmentation.
Figure 3: Manual disassembly of e-waste dataset.
Figure 4: SAM2 instance segmentation for logic board.
Figure 5: SAM2 instance segmentation for logic board attached at a laptop.
Figure 6: SAM2 instance segmentation for logic board, heat sink and IO board.
Figure 7: SAM2 instance segmentation for double-fan. All images use high-resolution input by default unless otherwise noted.
Figure 8: YOLOv8 training and validation loss.
Figure 9: YOLOv8 instance segmentation for four components.
ModelmAP 50mAP 50-95
SAM28.4%9%
YOLOv898.8%85%
Table 2: Segmentation results of two models
More in this topic

Instance Segmentation

Instance segmentation identifies each object and assigns it a separate pixel-level mask in images or video.

See all 159 papers

Explore similar work

CardsList

  1. Don't waste SAM

    Jun 9, 2026Nermeen Abou Baker, Uwe HandmannSupervised Fine-TuningPromptable Image Segmentation

    1- Ruhr West University of Applied Sciences - Dept of Computer Science Lutzowstrassse 5, 46236 Bottrop - Germany

  2. A.R.I.S.: Automated Recycling Identification System for E-Waste Classification Using Deep Learning

    Feb 19, 2026Dhruv Talwar, Harsh Desai, Wendong Yin +2Classification

    Apple

  3. GeoSAM-Lite: A Lightweight Foundation Model for Onboard Remote Sensing Segmentation

    Jul 4, 2026Yongcong Wang, Jie Zhang, Rui Jiang +3Remote Sensing Image SegmentationGeospatial Foundation Models

    College of Telecommunications and Information Engineering, Nanjing University of Posts and Telecommunications, Nanjing 210003, China