pith. machine review for the scientific record. sign in

arxiv: 2604.17231 · v1 · submitted 2026-04-19 · 💻 cs.CV · cs.RO

Recognition: unknown

Fringe Projection Based Vision Pipeline for Autonomous Hard Drive Disassembly

Badrinath Balasubramaniam , Vignesh Suresh , Benjamin Metcalf , Beiwen Li
Authors on Pith no claims yet

Pith reviewed 2026-05-10 06:53 UTC · model grok-4.3

classification 💻 cs.CV cs.RO
keywords fringe projection profilometryinstance segmentationdepth completionrobotic disassemblyhard disk drives3D sensinge-waste recyclingcomputer vision
0
0 comments X

The pith

A single fringe projection camera-projector pair supplies pixel-aligned depth maps and instance masks for robotic hard drive disassembly without separate registration steps.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper develops an integrated vision pipeline to support robotic disassembly of hard disk drives, a key step in recovering value from e-waste. It combines fringe projection profilometry for high-resolution 3D sensing with selective depth completion only where the projector fails and a lightweight segmentation network to identify components such as platters and fasteners. Because the same optical setup produces both the depth data and the segmentation input, the resulting 3D geometry and semantic labels remain perfectly aligned at every pixel. The system reports strong quantitative results on accuracy and speed while using simulation data to enlarge the training set. This matters for robotics because downstream manipulation tasks need reliable spatial and semantic information that does not require additional calibration between sensors.

Core claim

The authors establish that fringe projection profilometry can serve simultaneously as the primary depth sensor and as the imaging source for a real-time instance segmentation network. Selective triggering of a learned depth-completion model (Depth Anything V2 Base) fills gaps only when the fringe pattern fails, and the entire stack is optimized for deployment. On the evaluation data this yields box mAP@50 of 0.960 and mask mAP@50 of 0.957 for segmentation, depth RMSE of 2.317 mm, and a combined latency of 12.86 ms (77.7 FPS) for the platter-facing inference path. The approach avoids the registration overhead of typical RGB-D pipelines and is augmented by sim-to-real transfer learning.

What carries the argument

Fringe Projection Profilometry (FPP) module that re-uses the same camera-projector hardware for both depth sensing and instance segmentation, with selective depth completion triggered only on FPP failure regions.

If this is right

  • Pixel-wise alignment between depth and segmentation masks removes the need for extrinsic calibration between separate sensors.
  • Real-time performance at 77.7 FPS supports closed-loop control of robotic arms during disassembly.
  • High segmentation accuracy (0.96 mAP) enables precise localization of small fasteners and platters.
  • The public synthetic HDD dataset can reduce data-collection costs for similar recycling tasks.
  • Selective triggering limits the computational cost of depth completion to only the regions where FPP fails.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same optical architecture could be adapted to other small electronic devices if the fringe pattern density and segmentation classes are re-tuned.
  • Robustness testing on drives with reflective coatings or heavy wear would be a direct next measurement to confirm the weakest assumption.
  • End-to-end integration with force-controlled grippers could be tested by feeding the aligned 3D masks directly into motion planning.

Load-bearing premise

That selective depth completion can be triggered reliably across real-world variations in drive models, lighting, and surface conditions without introducing systematic geometric errors.

What would settle it

Deploy the pipeline on a fresh collection of hard drives never seen in training or simulation, under changed ambient lighting and surface finishes, and check whether segmentation mAP@50 falls below 0.90 or depth RMSE exceeds 5 mm.

Figures

Figures reproduced from arXiv: 2604.17231 by Badrinath Balasubramaniam, Beiwen Li, Benjamin Metcalf, Vignesh Suresh.

Figure 1
Figure 1. Figure 1: outlines the complete algorithm. Trained Neural Network Instance Segmentation Real World Data Segmentation Masks and Bounding Boxes Logic-Driven State Recognition Depth Map from FPP+MMDC-net Platter Facing Camera PCB Facing Camera Synthetic Data Train Fine Tune Digital Twin Real World System Inference 3D Reconstruction +Segmentation Masks Depth Map from FPP [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Representative HDD component taxonomy used for instance segmentation. Left: Platter Facing view showing the internal mechanical components after removal of the top cover. Right: PCB Facing view showing the external electronics side of the drive. Colors denote the 11 component classes used in the proposed semantic segmentation framework, spanning mechanical and moving parts, electronics and interfaces, and … view at source ↗
Figure 3
Figure 3. Figure 3: Schematic of YOLOv11n-seg (Re-used from [28]) and triangulation are sufficient to generate a high-fidelity point cloud with sub-millimeter accuracy. Conversely, the Platter Facing state exposes the mirror-like central platter, and other metallic and optically challenging components that cause saturation and phase unwrapping failures in structured light systems. Upon detecting this state, the pipeline selec… view at source ↗
Figure 4
Figure 4. Figure 4: Schematic of FPP (Adapted from [29]) formulation: 𝜙(𝑥, 𝑦) = − arctan( ∑18 𝑛=1 𝐼𝑛 sin 𝛿𝑛 ∑18 𝑛=1 𝐼𝑛 cos 𝛿𝑛 ) . (3) We unwrap this phase using the graycoding technique [30] to generate the absolute phase. Using the system calibration parameters derived from Li et al. [31], we map the absolute phase and camera coordinates (𝑢𝑐 , 𝑣𝑐 ) to the world coordinates(𝑥𝑤, 𝑦𝑤, 𝑧𝑤) via triangulation, generating the dense … view at source ↗
Figure 5
Figure 5. Figure 5: Real-World Hardware Set Up (Adapted from [32]) Calibration: System calibration is performed using a 5×9 circle grid target using the technique proposed in Li et al [31]. It involves rotating the circle board through 18 different poses and capturing 52 images of horizontal and vertical fringe patterns. The horizontal and vertical phase maps computed are then used in conjunction with OpenCV’s camera calibrat… view at source ↗
Figure 6
Figure 6. Figure 6: Synthetic Data Input and Ground Truth Pairs with (i) Sample 2.5 inch HDD (front), (ii) Sample 3.5 inch HDD (front), and (iii) Sample 3.5 inch HDD (back) We utilize 14 CAD models sourced from SketchFab [36] and GrabCAD [37]. The projector from our digital twin illuminates the object with white light. We rotate each CAD model through multiple orientations. At each orientation, we capture a projector illumina… view at source ↗
Figure 7
Figure 7. Figure 7: Qualitative instance segmentation results on representative unseen HDD test samples. Rows (a)–(c) show three different HDD examples spanning both Platter Facing and PCB Facing orientations. Columns show the projector-illuminated input image, the YOLOv11n-seg prediction, and the corresponding ground-truth annotation. The results illustrate accurate localization and boundary delineation of both macro-compone… view at source ↗
Figure 8
Figure 8. Figure 8: Baseline FPP reconstruction for a PCB Facing HDD. (a) Representative fringe-pattern image acquired from the PCB Facing side of the drive. (b) Corresponding 3D reconstruction obtained using standard FPP, with color indicating depth 𝑧 in millimeters. Because the exposed PCB side is largely matte and contains few highly specular regions, baseline FPP is sufficient to recover the 3D profiles of the PCB and ass… view at source ↗
Figure 9
Figure 9. Figure 9: Baseline FPP reconstruction for a Platter Facing HDD. (a) Fringe-pattern image acquired from the Platter Facing side of the drive. (b) Corresponding 3D reconstruction obtained using standard FPP, with color indicating depth z in millimeters. Strong specular reflections from the platter and reflective metallic regions lead to sparse reconstruction and missing depth, motivating the use of MMDC-Net in the pro… view at source ↗
Figure 10
Figure 10. Figure 10: Qualitative comparison of MMDC-Net variants trained with different Depth Anything V2 backbones on a representative Platter Facing HDD. Rows (i)–(iii) correspond to Depth Anything V2 Large, Base, and Small, respectively. Columns show (a) ground-truth depth, (b) final reconstruction obtained by combining reliable FPP depth with MMDC-Net predictions, (c) MMDC-Net prediction in unreliable regions only, (d) gr… view at source ↗
Figure 11
Figure 11. Figure 11: Integrated semantic and geometric output of the proposed vision pipeline for representative Platter Facing and PCB Facing HDDs. Rows (a) and (b) correspond to the Platter Facing and PCB Facing cases, respectively. Within each row, the columns show: the predicted instance segmentation masks overlaid on the projector-illuminated input image, the reconstructed 3D profile obtained from the adaptive FPP pipeli… view at source ↗
read the original abstract

Unrecovered e-waste represents a significant economic loss. Hard disk drives (HDDs) comprise a valuable e-waste stream necessitating robotic disassembly. Automating the disassembly of HDDs requires holistic 3D sensing, scene understanding, and fastener localization, however current methods are fragmented, lack robust 3D sensing, and lack fastener localization. We propose an autonomous vision pipeline which performs 3D sensing using a Fringe Projection Profilometry (FPP) module, with selective triggering of a depth completion module where FPP fails, and integrates this module with a lightweight, real-time instance segmentation network for scene understanding and critical component localization. By utilizing the same FPP camera-projector system for both our depth sensing and component localization modules, our depth maps and derived 3D geometry are inherently pixel-wise aligned with the segmentation masks without registration, providing an advantage over RGB-D perception systems common in industrial sensing. We optimize both our trained depth completion and instance segmentation networks for deployment-oriented inference. The proposed system achieves a box mAP@50 of 0.960 and mask mAP@50 of 0.957 for instance segmentation, while the selected depth completion configuration with the Depth Anything V2 Base backbone achieves an RMSE of 2.317 mm and MAE of 1.836 mm; the Platter Facing learned inference stack achieved a combined latency of 12.86 ms and a throughput of 77.7 Frames Per Second (FPS) on the evaluation workstation. Finally, we adopt a sim-to-real transfer learning approach to augment our physical dataset. The proposed perception pipeline provides both high-fidelity semantic and spatial data which can be valuable for downstream robotic disassembly. The synthetic dataset developed for HDD instance segmentation will be made publicly available.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 3 minor

Summary. The manuscript presents a vision pipeline for autonomous hard disk drive (HDD) disassembly that integrates Fringe Projection Profilometry (FPP) for primary 3D sensing, with selective triggering of monocular depth completion (Depth Anything V2 Base) on FPP failure, and a lightweight real-time instance segmentation network for scene understanding and fastener/component localization. The approach exploits the shared camera-projector hardware to achieve inherent pixel-wise alignment between depth maps and segmentation masks without registration. It reports box mAP@50 of 0.960 and mask mAP@50 of 0.957 for segmentation, depth RMSE of 2.317 mm and MAE of 1.836 mm, combined latency of 12.86 ms and 77.7 FPS for the platter-facing stack, and employs sim-to-real transfer to augment the physical dataset, with plans to release the synthetic data publicly.

Significance. If the hybrid pipeline and reported metrics hold, the work offers a practical contribution to robotic e-waste recycling by delivering an integrated, aligned semantic-spatial perception system optimized for real-time deployment. The explicit quantitative results on segmentation accuracy, depth precision, and inference throughput, combined with the public dataset release, provide concrete value for downstream robotics tasks. The emphasis on hardware-level alignment and inference optimization are notable strengths.

major comments (2)
  1. [Method section (hybrid depth pipeline description)] The selective triggering of depth completion is load-bearing for the hybrid 3D sensing claims and the reported RMSE/MAE values, yet the manuscript provides no explicit decision rule, threshold value, false-positive/negative rates, or validation experiments quantifying depth errors introduced on real HDDs under varying reflectivity, lighting, surface wear, or model diversity. Without this, the end-to-end accuracy and robustness of the pipeline remain unsubstantiated.
  2. [Experiments (sim-to-real and dataset)] The sim-to-real transfer approach for instance segmentation is central to the generalization and dataset augmentation claims, but no domain-gap metrics, cross-model hold-out results on physical data from different HDD models, or quantitative comparison of synthetic vs. real performance are reported. This leaves the effectiveness of the transfer unquantified and risks overstatement of the mAP figures.
minor comments (3)
  1. [Abstract and Experiments] Dataset size, number of HDD models, diversity of conditions, and train/test split details are not provided, which would better contextualize the mAP@50, RMSE, and FPS metrics.
  2. [Experiments] Baseline comparisons to alternative depth sensing (e.g., standard RGB-D or other completion methods) or segmentation networks are absent, limiting assessment of the advance over the fragmented methods noted in the introduction.
  3. [Experiments] Failure-case analysis for the full pipeline (e.g., when FPP fails and completion is triggered) would strengthen the practical claims.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We are grateful to the referee for their insightful comments, which have helped us identify areas for improvement in the manuscript. We provide point-by-point responses to the major comments below.

read point-by-point responses

  1. Referee: [Method section (hybrid depth pipeline description)] The selective triggering of depth completion is load-bearing for the hybrid 3D sensing claims and the reported RMSE/MAE values, yet the manuscript provides no explicit decision rule, threshold value, false-positive/negative rates, or validation experiments quantifying depth errors introduced on real HDDs under varying reflectivity, lighting, surface wear, or model diversity. Without this, the end-to-end accuracy and robustness of the pipeline remain unsubstantiated.

    Authors: We thank the referee for highlighting this important aspect. Upon review, we recognize that the manuscript lacks an explicit description of the decision rule for selective triggering of the depth completion module. The triggering is intended to activate when FPP fails, but specific thresholds and validation were omitted. To address this, we will update the Method section with a detailed explanation of the decision criteria, including threshold values and metrics used to detect FPP failure. Additionally, we will incorporate validation experiments on real HDDs under varied conditions (reflectivity, lighting, surface wear, model diversity) and report the associated false-positive and false-negative rates for the triggering logic. This will substantiate the end-to-end accuracy and robustness of the hybrid pipeline. revision: yes

  2. Referee: [Experiments (sim-to-real and dataset)] The sim-to-real transfer approach for instance segmentation is central to the generalization and dataset augmentation claims, but no domain-gap metrics, cross-model hold-out results on physical data from different HDD models, or quantitative comparison of synthetic vs. real performance are reported. This leaves the effectiveness of the transfer unquantified and risks overstatement of the mAP figures.

    Authors: We appreciate the referee's point regarding the sim-to-real transfer. The current manuscript reports performance on the augmented dataset but does not provide domain-gap metrics, cross-model hold-out results, or direct comparisons of synthetic versus real performance. We will revise the Experiments section to include these quantitative analyses, such as domain adaptation metrics, performance on hold-out physical data from different HDD models, and ablation studies comparing models trained with and without the synthetic data. This will better quantify the effectiveness of the transfer learning approach and support the reported mAP values. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical pipeline with direct metric measurements

full rationale

The paper describes an integrated hardware-software vision system evaluated via standard held-out test metrics (box/mask mAP@50, RMSE/MAE in mm, latency/FPS). No equations, derivations, or 'predictions' are presented that reduce to fitted inputs by construction. Selective triggering and sim-to-real augmentation are described as engineering choices without self-referential definitions or load-bearing self-citations that close the loop on the reported numbers. All performance figures are direct empirical outcomes on test data, not re-expressions of training objectives or prior author results.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

The central claims rest on standard computer-vision assumptions about surface reflectance for FPP and the generalization properties of trained neural networks; no new entities are postulated.

free parameters (2)
  • depth-completion triggering threshold
    Condition for switching from FPP to neural depth completion is not specified and must be tuned for the target environment.
  • network training hyperparameters
    Standard deep-learning hyperparameters for Depth Anything V2 and the segmentation model are fitted during training.
axioms (2)
  • domain assumption FPP yields accurate depth maps when projected patterns are reliably captured by the camera
    Invoked to justify selective use of the FPP module.
  • domain assumption Sim-to-real transfer improves robustness of segmentation and depth models on physical HDDs
    Used to augment the physical training set.

pith-pipeline@v0.9.0 · 5631 in / 1464 out tokens · 42440 ms · 2026-05-10T06:53:45.289789+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

44 extracted references · 34 canonical work pages · 2 internal anchors

  1. [1]

    C. P. Baldé, R. Kuehr, T. Yamamoto, R. McDonald, E. D’Angelo, S. Althaf, G. Bel, O. Deubzer, E. Fernandez-Cubillo, V. Forti, V. Gray, S. Herat, S. Honda, G. Iattoni, D. S. Khetriwal, V. Luda di Cortemiglia, Y. Lobuntsova, I. Nnorom, N. Pralat, M. Wagner, The global e-waste monitor 2024, Tech. rep., United Nations Institute for Training and Research (UNITA...

    2024

  2. [2]

    Sabbaghi, W

    M. Sabbaghi, W. Cade, W. Olson, S. Behdad, The Global Flow of Hard Disk Drives: Quantifying the Concept of Value Leakage in E-waste RecoverySystems,JournalofIndustrialEcology23(3)(2019)560–573,_eprint:https://onlinelibrary.wiley.com/doi/pdf/10.1111/jiec.12765. doi:10.1111/jiec.12765. URLhttps://onlinelibrary.wiley.com/doi/abs/10.1111/jiec.12765

    work page doi:10.1111/jiec.12765 2019

  3. [3]

    URLhttps://doi.org/10.1021/acs.est.5b02264

    K.Habib,K.Parajuly,H.Wenzel,TrackingtheFlowofResourcesinElectronicWaste-TheCaseofEnd-of-LifeComputerHardDiskDrives, Environmental Science & Technology 49 (20) (2015) 12441–12449.doi:10.1021/acs.est.5b02264. URLhttps://doi.org/10.1021/acs.est.5b02264

    work page doi:10.1021/acs.est.5b02264 2015

  4. [4]

    URLhttps://www.sciencedirect.com/science/article/pii/S0959652619342477

    H.Tanvar,A.Barnwal,N.Dhawan,Characterizationandevaluationofdiscardedharddiscdrivesforrecoveryofcopperandrareearthvalues, Journal of Cleaner Production 249 (2020) 119377.doi:10.1016/j.jclepro.2019.119377. URLhttps://www.sciencedirect.com/science/article/pii/S0959652619342477

    work page doi:10.1016/j.jclepro.2019.119377 2020

  5. [5]

    K. Liu, Q. Tan, J. Yu, M. Wang, A global perspective on e-waste recycling, Circular Economy 2 (1) (2023) 100028.doi:10.1016/ j.cec.2023.100028. URLhttps://www.sciencedirect.com/science/article/pii/S2773167723000055

    work page arXiv 2023

  6. [6]

    Barman, S

    T. Barman, S. Coleman, D. Kerr, S. Harrigan, J. Quinn, Optimizing Industrial E-Waste Recycling with Attention-Driven Deep Learning for PCB Segmentation Using Hyperspectral Imaging, in: 2025 IEEE Symposia on Computational Intelligence for Energy, Transport and Environmental Sustainability (CIETES), 2025, pp. 1–7.doi:10.1109/CIETES63869.2025.10995066. URLht...

    work page doi:10.1109/cietes63869.2025.10995066 2025

  7. [7]

    A. F. Hussein, W. M. Hamanah, M. A. Abido, Harnessing hyperspectral imaging and deep learning for advanced e-waste classification using three spectral bands, Results in Engineering 27 (2025) 106110.doi:10.1016/j.rineng.2025.106110. URLhttps://www.sciencedirect.com/science/article/pii/S2590123025021826

    work page doi:10.1016/j.rineng.2025.106110 2025

  8. [8]

    Picon, P

    A. Picon, P. Galan, A. Bereciartua-Perez, L. Benito-del Valle, On the analysis of adapting deep learning methods to hyperspectral imaging. Use case for WEEE recycling and dataset, Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy 330 (2025) 125665. doi:10.1016/j.saa.2024.125665. URLhttps://www.sciencedirect.com/science/article/pii/S13861...

    work page doi:10.1016/j.saa.2024.125665 2025

  9. [9]

    Mohsin, S

    M. Mohsin, S. Rovetta, F. Masulli, A. Cabri, Heatmap Visualization for Deep Learning Analysis of Waste Printed Circuit Boards, in: 2025 IEEE 2nd International Conference on Electronics, Communications and Intelligent Science (ECIS), 2025, pp. 1–5.doi:10.1109/ ECIS65594.2025.11086768. URLhttps://ieeexplore.ieee.org/document/11086768

    work page arXiv 2025

  10. [10]

    R. R. Selvaraju, M. Cogswell, A. Das, R. Vedantam, D. Parikh, D. Batra, Grad-CAM: Visual Explanations from Deep Networks via Gradient-Based Localization, in: 2017 IEEE International Conference on Computer Vision (ICCV), 2017, pp. 618–626, iSSN: 2380-7504. doi:10.1109/ICCV.2017.74. URLhttps://ieeexplore.ieee.org/document/8237336

    work page doi:10.1109/iccv.2017.74 2017

  11. [11]

    Ahmed, D

    S. Ahmed, D. F. Sanam, Exploration of an E-waste Prediction Model for Collection System Optimization Using Deep Learning Models in E-waste Management Facilities, in: 2025 2nd International Conference on Next-Generation Computing, IoT and Machine Learning (NCIM), 2025, pp. 1–6.doi:10.1109/NCIM65934.2025.11160027. URLhttps://ieeexplore.ieee.org/abstract/doc...

    work page doi:10.1109/ncim65934.2025.11160027 2025

  12. [12]

    Jahanian, Q

    A. Jahanian, Q. H. Le, K. Youcef-Toumi, D. Tsetserukou, See the E-Waste! Training Visual Intelligence to See Dense Circuit Boards for Recycling, 2019, pp. 0–0. URLhttps://openaccess.thecvf.com/content_CVPRW_2019/html/cv4gc/Jahanian_See_the_E- Waste_Training_Visual_Intelligence_to_See_Dense_Circuit_CVPRW_2019_paper.html

    2019

  13. [13]

    C. Liu, B. Balasubramaniam, N. A. Yancey, M. H. Severson, A. Shine, P. Bove, B. Li, X. Liang, M. Zheng, RAISE: A Robot-Assisted SelectIve Disassembly and Sorting System for End-of-Life Phones, Resources, Conservation and Recycling 225 (2026) 108609.doi: 10.1016/j.resconrec.2025.108609. URLhttps://www.sciencedirect.com/science/article/pii/S0921344925004860

    work page doi:10.1016/j.resconrec.2025.108609 2026

  14. [14]

    G. Foo, S. Kara, M. Pagnucco, Screw detection for disassembly of electronic waste using reasoning and re-training of a deep learning model, Procedia CIRP 98 (2021) 666–671.doi:10.1016/j.procir.2021.01.172. URLhttps://www.sciencedirect.com/science/article/pii/S2212827121002031

    work page doi:10.1016/j.procir.2021.01.172 2021

  15. [15]

    Karbouj, G

    B. Karbouj, G. A. Topalian-Rivas, J. Krüger, Comparative Performance Evaluation of One-Stage and Two-Stage Object Detectors for Screw Head Detection and Classification in Disassembly Processes, Procedia CIRP 122 (2024) 527–532.doi:10.1016/j.procir.2024.01.077. URLhttps://www.sciencedirect.com/science/article/pii/S2212827124001021

    work page doi:10.1016/j.procir.2024.01.077 2024

  16. [16]

    Puttero, A

    S. Puttero, A. Nassehi, E. Verna, G. Genta, M. Galetto, Automatic object detection for disassembly and recycling of electronic board components, Procedia CIRP 127 (2024) 206–211.doi:10.1016/j.procir.2024.07.036. URLhttps://www.sciencedirect.com/science/article/pii/S2212827124003457

    work page doi:10.1016/j.procir.2024.07.036 2024

  17. [17]

    D. P. Brogan, N. M. DiFilippo, M. K. Jouaneh, Deep learning computer vision for robotic disassembly and servicing applications, Array 12 (2021) 100094.doi:10.1016/j.array.2021.100094. URLhttps://www.sciencedirect.com/science/article/pii/S2590005621000394

    work page doi:10.1016/j.array.2021.100094 2021

  18. [18]

    H. Zhao, C. Liu, B. Balasubramaniam, J. Li, J. Song, X. Liang, M. Zheng, B. Li, Precision 3D profilometry of consumer-grade computer enclosures using high dynamic range fringe projection, Applied Optics 64 (30) (2025) 8986–8994.doi:10.1364/AO.575966. URLhttps://opg.optica.org/ao/abstract.cfm?uri=ao-64-30-8986

    work page doi:10.1364/ao.575966 2025

  19. [19]

    Yildiz, T

    E. Yildiz, T. Brinker, E. Renaudo, J. Hollenstein, S. Haller-Seeber, J. Piater, F. Wörgötter, A Visual Intelligence Scheme for Hard Drive DisassemblyinAutomatedRecyclingRoutines:,in:ProceedingsoftheInternationalConferenceonRobotics,ComputerVisionandIntelligent Systems, SCITEPRESS - Science and Technology Publications, Budapest, Hungary, 2020, pp. 17–27.do...

    work page doi:10.5220/0010016000170027 2020

  20. [20]

    Yildiz, E

    E. Yildiz, E. Renaudo, J. Hollenstein, J. Piater, F. Wörgötter, An Extended Visual Intelligence Scheme for Disassembly in Automated RecyclingRoutines,in:P.Galambos,E.Kayacan,K.Madani(Eds.),Robotics,ComputerVisionandIntelligentSystems,SpringerInternational Publishing, Cham, 2022, pp. 25–50.doi:10.1007/978-3-031-19650-8_2

    work page doi:10.1007/978-3-031-19650-8_2 2022

  21. [21]

    Yildiz, F

    E. Yildiz, F. Wörgötter, DCNN-Based Screw Detection for Automated Disassembly Processes, in: 2019 15th International Conference on Signal-Image Technology & Internet-Based Systems (SITIS), 2019, pp. 187–192.doi:10.1109/SITIS.2019.00040. URLhttps://ieeexplore.ieee.org/document/9067965

    work page doi:10.1109/sitis.2019.00040 2019

  22. [22]

    Rojas, A

    C. Rojas, A. Rodríguez-Sánchez, E. Renaudo, Efficient Segmentation of E-Waste Devices With Deep Learning for Robotic Recycling, in: Emerging Topics in Pattern Recognition and Artificial Intelligence, Vol. Volume 9 of Series on Language Processing, Pattern Recognition, and Intelligent Systems, WORLD SCIENTIFIC, 2023, pp. 123–144.doi:10.1142/9789811289125_0...

    work page doi:10.1142/9789811289125_0005 2023

  23. [23]

    El Yacoubi, E

    C.Rojas,A.Rodríguez-Sánchez,E.Renaudo,DeepLearningforFastSegmentationofE-wasteDevices’InnerPartsinaRecyclingScenario, in: M. El Yacoubi, E. Granger, P. C. Yuen, U. Pal, N. Vincent (Eds.), Pattern Recognition and Artificial Intelligence, Springer International Publishing, Cham, 2022, pp. 161–172.doi:10.1007/978-3-031-09037-0_14

    work page doi:10.1007/978-3-031-09037-0_14 2022

  24. [24]

    B. Li, High-speed 3D optical sensing for manufacturing research and industrial sensing applications, Transactions on Energy Systems and Engineering Applications 3 (2) (2022) 1–12.doi:10.32397/tesea.vol3.n2.490. URLhttps://revistas.utb.edu.co/tesea/article/view/490

    work page doi:10.32397/tesea.vol3.n2.490 2022

  25. [25]

    Balasubramaniam, V

    B. Balasubramaniam, V. Suresh, Y. Cheng, J. Li, B. Li, Application-driven multi-modal depth completion in fringe projection profilometry, Optics and Lasers in Engineering 200 (2026) 109587.doi:10.1016/j.optlaseng.2025.109587. URLhttps://www.sciencedirect.com/science/article/pii/S0143816625007717

    work page doi:10.1016/j.optlaseng.2025.109587 2026

  26. [26]

    URLhttps://imagine-h2020.eu/hdd-taxonomy.php

    IMAGINE. URLhttps://imagine-h2020.eu/hdd-taxonomy.php

  27. [27]

    YOLOv11: An Overview of the Key Architectural Enhancements

    R. Khanam, M. Hussain, YOLOv11: An Overview of the Key Architectural Enhancements, arXiv:2410.17725 [cs] version: 1 (Oct. 2024). doi:10.48550/arXiv.2410.17725. URLhttp://arxiv.org/abs/2410.17725

    work page internal anchor Pith review doi:10.48550/arxiv.2410.17725 2024

  28. [28]

    Q. Wang, X. Wei, K. Li, B. Cao, W. Zhang, Non-Contact Measurement of Sunflower Flowerhead Morphology Using Mobile- Boosted Lightweight Asymmetric (MBLA)-YOLO and Point Cloud Technology, Agriculture 15 (21) (2025) 2180.doi:10.3390/ agriculture15212180. URLhttps://www.mdpi.com/2077-0472/15/21/2180

    2025

  29. [29]

    URLhttps://www.mdpi.com/2077-0472/15/15/1702

    L.H.Naqvi,B.Balasubramaniam,J.Li,L.Liu,B.Li,Four-DimensionalHyperspectralImagingforFruitandVegetableGrading,Agriculture 15 (15) (2025) 1702.doi:10.3390/agriculture15151702. URLhttps://www.mdpi.com/2077-0472/15/15/1702

    work page doi:10.3390/agriculture15151702 2025

  30. [30]

    Zhang, High-Speed 3D Imaging with Digital Fringe Projection Techniques, CRC Press, 2018, google-Books-ID: p22mCwAAQBAJ

    S. Zhang, High-Speed 3D Imaging with Digital Fringe Projection Techniques, CRC Press, 2018, google-Books-ID: p22mCwAAQBAJ

    2018

  31. [31]

    B. Li, N. Karpinsky, S. Zhang, Novel calibration method for structured-light system with an out-of-focus projector, Applied Optics 53 (16) (2014) 3415–3426, publisher: Optica Publishing Group.doi:10.1364/AO.53.003415. URLhttps://opg.optica.org/ao/abstract.cfm?uri=ao-53-16-3415 Balasubramaniam et al.:Preprint submitted to ElsevierPage 19 of 20 Fringe Proje...

    work page doi:10.1364/ao.53.003415 2014

  32. [32]

    URLhttps://www.mdpi.com/2079-9292/12/4/859

    B.Balasubramaniam,J.Li,L.Liu,B.Li,3DImagingwithFringeProjectionforFoodandAgriculturalApplications—ATutorial,Electronics 12 (4) (2023) 859, number: 4.doi:10.3390/electronics12040859. URLhttps://www.mdpi.com/2079-9292/12/4/859

    work page doi:10.3390/electronics12040859 2023

  33. [33]

    Balasubramaniam, B

    B. Balasubramaniam, B. Li, Single Shot 3D Shape Measurement of Non-Volatile Data Storage Devices, Vol. Volume 2: Manufacturing EquipmentandAutomation;ManufacturingProcesses;ManufacturingSystems;Nano/Micro/MesoManufacturing;QualityandReliabilityof InternationalManufacturingScienceandEngineeringConference,2023,_eprint:https://asmedigitalcollection.asme.org/...

    work page doi:10.1115/msec2023-104380 2023

  34. [34]

    Dwyer, J

    B. Dwyer, J. Nelson, T. Hansen, et al., Roboflow (Version 1.0) [Software], computer vision (2026). URLhttps://roboflow.com

    2026

  35. [35]

    Zheng, S

    Y. Zheng, S. Wang, Q. Li, B. Li, Fringe projection profilometry by conducting deep learning from its digital twin, Optics Express 28 (24) (2020) 36568–36583.doi:10.1364/OE.410428. URLhttps://opg.optica.org/oe/abstract.cfm?uri=oe-28-24-36568

    work page doi:10.1364/oe.410428 2020

  36. [36]

    URLhttps://sketchfab.com

    Sketchfab - The best 3D viewer on the web. URLhttps://sketchfab.com

  37. [37]

    URLhttps://grabcad.com/

    GrabCAD Making Additive Manufacturing at Scale Possible. URLhttps://grabcad.com/

  38. [38]

    SAM 3: Segment Anything with Concepts

    N. Carion, L. Gustafson, Y.-T. Hu, S. Debnath, R. Hu, D. Suris, C. Ryali, K. V. Alwala, H. Khedr, A. Huang, et al., Sam 3: Segment anything with concepts, arXiv preprint arXiv:2511.16719 (2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  39. [39]

    F. Lin, J. Yuan, S. Wu, F. Wang, Z. Wang, UniNeXt: Exploring A Unified Architecture for Vision Recognition, in: Proceedings of the 31st ACMInternationalConferenceonMultimedia,MM’23,AssociationforComputingMachinery,NewYork,NY,USA,2023,pp.3200–3208. doi:10.1145/3581783.3612260. URLhttps://dl.acm.org/doi/10.1145/3581783.3612260

    work page doi:10.1145/3581783.3612260 2023

  40. [40]

    G. P. B, R. Marion Lincy G, A. Rishekeeshan, Deekshitha, Accelerating Native Inference Model Performance in Edge Devices using TensorRT, in: 2024 IEEE Recent Advances in Intelligent Computational Systems (RAICS), 2024, pp. 1–7, iSSN: 2769-5565.doi: 10.1109/RAICS61201.2024.10690032. URLhttps://ieeexplore.ieee.org/abstract/document/10690032

    work page doi:10.1109/raics61201.2024.10690032 2024

  41. [41]

    URLhttps://onnx.ai/onnx/intro/

    Introduction to ONNX - ONNX 1.21.0 documentation. URLhttps://onnx.ai/onnx/intro/

  42. [42]

    URLhttps://developer.nvidia.com/tensorrt

    NVIDIA TensorRT. URLhttps://developer.nvidia.com/tensorrt

  43. [43]

    2011–2018.doi:10.1109/HPCC-DSS- SmartCity-DependSys57074.2022.00299

    Y.Zhou,K.Yang,ExploringTensorRTtoImproveReal-TimeInferenceforDeepLearning,in:2022IEEE24thIntConfonHighPerformance Computing & Communications; 8th Int Conf on Data Science & Systems; 20th Int Conf on Smart City; 8th Int Conf on Dependability in Sensor, Cloud & Big Data Systems & Application (HPCC/DSS/SmartCity/DependSys), 2022, pp. 2011–2018.doi:10.1109/HP...

    work page doi:10.1109/hpcc-dss- 2022

  44. [44]

    Jafarpourmarzouni, S

    Sumaiya, R. Jafarpourmarzouni, S. Lu, Z. Dong, Enhancing Real-time Inference Performance for Time-Critical Software-Defined Vehicles, in: 2024 IEEE International Conference on Mobility, Operations, Services and Technologies (MOST), 2024, pp. 101–113.doi:10.1109/ MOST60774.2024.00019. URLhttps://ieeexplore.ieee.org/abstract/document/10607059 Balasubramania...

    work page arXiv 2024