Industrial Part Detection based on CAD Model Matching and Appearance–Silhouette Score Fusion

Authors

  • Minglei Zhu School of Mechanical Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China
  • Junjie Shen School of Mechanical Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China
  • Junjie Ma VEM China Co., Ltd., Wuxi 214100, China

DOI:

https://doi.org/10.54691/edbn3q81

Keywords:

Industrial Part Detection; CAD Model Matching; Confidence Score Fusion; Silhouette Similarity; Unseen-Object Detection.

Abstract

Part detectors based on computer-aided design (CAD) models must rank candidate regions by confidence. To assess the complementary value of CAD silhouette information and its effect on proposal ranking and industrial part detection, this study combines a library-wide mean silhouette score with appearance confidence. Experiments use Segment Anything Model (SAM) proposals and DINOv2 appearance features on XYZIBD RGB images. An internally pre-specified development rule selected appearance scoring because of its higher coverage AUC (0.645 versus 0.620), making the conditional detector identical to the appearance baseline. On the held-out set, pre-specified unit-weight silhouette addition increased the class-agnostic average precision (CAAP) point estimate by 1.28 percentage points from a 1.25% baseline, with a 95% difference interval of −0.20 to 2.68. Adding SAM-predicted IoU yielded a difference of −0.33 points, with an interval of −1.22 to 0.13. Instance coverage was 17.3% at IoU ≥ 0.5, defining the range of instances recoverable through rescoring. Post hoc rescoring of saved CNOS proposals yielded CAAP differences of +2.91 and −0.01 points for the study and native CNOS scorers, respectively. Classwise AP increased slightly for the study scorer and decreased for the native scorer. These findings support evaluating appearance and silhouette score fusion within complete detection pipelines while accounting for instance coverage, scoring configuration, and classwise AP.

Downloads

Download data is not yet available.

References

[1] Hinterstoisser, S., et al. (2012). Gradient response maps for real-time detection of texture-less objects. IEEE Transactions on Pattern Analysis and Machine Intelligence, 34(5), 876–888.

[2] Borgefors, G. (1988). Hierarchical chamfer matching: A parametric edge matching algorithm. IEEE Transactions on Pattern Analysis and Machine Intelligence, 10(6), 849–865.

[3] Nguyen, V. N., Groueix, T., Ponimatkin, G., Lepetit, V., & Hodaň, T. (2023). CNOS: A strong baseline for CAD-based novel object segmentation. In Proceedings of the IEEE/CVF International Conference on Computer Vision Workshops (pp. 2126–2132). Paris, France.

[4] Lin, J., Liu, L., Lu, D., & Jia, K. (2024). SAM-6D: Segment anything model meets zero-shot 6D object pose estimation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (pp. 27906–27916). Seattle, USA.

[5] Cho, S., Park, S., & Oh, I. (2026). MUSE: Model-based uncertainty-aware similarity estimation for zero-shot 2D object detection and segmentation. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (pp. 6090–6100). Tucson, USA.

[6] Hodaň, T., Haluza, P., Obdržálek, Š., Matas, J., Lourakis, M., & Zabulis, X. (2017). T-LESS: An RGB-D dataset for 6D pose estimation of texture-less objects. In Proceedings of the IEEE Winter Conference on Applications of Computer Vision (pp. 880–888). Santa Rosa, USA.

[7] Xiang, Y., Schmidt, T., Narayanan, V., & Fox, D. (2018). PoseCNN: A convolutional neural network for 6D object pose estimation in cluttered scenes. In Robotics: Science and Systems XIV. Pittsburgh, USA.

[8] Guyon, I., & Elisseeff, A. (2003). An introduction to variable and feature selection. Journal of Machine Learning Research, 3, 1157–1182.

[9] Lu, Y., P, J. J., Guo, Y., Ruozzi, N., & Xiang, Y. (2025). Adapting pre-trained vision models for novel instance detection and segmentation. In Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems (pp. 13341–13348). Hangzhou, China.

[10] Liu, S., et al. (2024). Grounding DINO: Marrying DINO with grounded pre-training for open-set object detection. In European Conference on Computer Vision (Vol. 15105, pp. 38–55). Lecture Notes in Computer Science. Milan, Italy.

[11] Gandyra, M., Santonicola, A., & Beetz, M. (2025). NOCTIS: Novel object cyclic threshold based instance segmentation. arXiv preprint arXiv:2507.01463. https://arxiv.org/abs/2507.01463

[12] Gandyra, M., Santonicola, A., & Beetz, M. (2025). Depth-NOCTIS: Depth-based novel object cyclic threshold based instance segmentation. In British Machine Vision Conference DIFA Workshop. Sheffield, UK.

[13] Ji, S., Kwon, Y., & Kim, H.-S. (2026). CLON: Cue-calibrated linguistic object onboarding for zero-shot 6D pose front-ends. arXiv preprint arXiv:2609.04784v1. https://arxiv.org/abs/2609.04784v1

[14] Alexe, B., Deselaers, T., & Ferrari, V. (2012). Measuring the objectness of image windows. IEEE Transactions on Pattern Analysis and Machine Intelligence, 34(11), 2189–2202.

[15] Guo, C., Pleiss, G., Sun, Y., & Weinberger, K. Q. (2017). On calibration of modern neural networks. In Proceedings of the 34th International Conference on Machine Learning (Vol. 70, pp. 1321–1330). Proceedings of Machine Learning Research. Sydney, Australia.

[16] Küppers, F., Kronenberger, J., Shantia, A., & Haselhoff, A. (2020). Multivariate confidence calibration for object detection. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (pp. 1322–1330). Virtual Conference.

[17] Hosang, J., Benenson, R., Dollár, P., & Schiele, B. (2016). What makes for effective detection proposals? IEEE Transactions on Pattern Analysis and Machine Intelligence, 38(4), 814–830.

[18] Hodaň, T., et al. (2024). BOP Challenge 2023 on detection, segmentation and pose estimation of seen and unseen rigid objects. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (pp. 5610–5619). Seattle, USA.

[19] Nguyen, V. N., et al. (2025). BOP Challenge 2024 on model-based and model-free 6D object pose estimation. arXiv preprint arXiv:2504.02812v4. https://arxiv.org/abs/2504.02812v4

[20] Huang, J., Hu, J., Yu, P. K. T., Ilic, S., Sundermeyer, M., & Busam, B. (2026). XYZ-IBD: Benchmarking robust 6D object pose estimation under real-world industrial complexity. arXiv preprint arXiv:2506.00599v3. https://arxiv.org/abs/2506.00599v3

[21] Hu, J., Xu, H., Huang, J., Yu, P. K. T., Ilic, S., & Busam, B. (2025). BioDet: Boosting industrial object detection with image preprocessing strategies. arXiv preprint arXiv:2510.21000. https://arxiv.org/abs/2510.21000

[22] Davis, J., & Goadrich, M. (2006). The relationship between precision-recall and ROC curves. In Proceedings of the 23rd International Conference on Machine Learning (pp. 233–240). Pittsburgh, USA.

[23] Lin, T.-Y., et al. (2014). Microsoft COCO: Common objects in context. In European Conference on Computer Vision (Vol. 8693, pp. 740–755). Lecture Notes in Computer Science. Zurich, Switzerland.

[24] COCO Dataset. (n.d.). pycocotools. https://github.com/cocodataset/cocoapi/blob/master/PythonAPI/pycocotools

[25] Hodaň, T. (n.d.). bop_toolkit. https://github.com/thodan/bop_toolkit

[26] BOP Benchmark. (n.d.). xyzibd dataset. https://huggingface.co/datasets/bop-benchmark/xyzibd

[27] Kirillov, A., et al. (2023). Segment anything. In Proceedings of the IEEE/CVF International Conference on Computer Vision (pp. 3992–4003). Paris, France.

[28] Oquab, M., et al. (2024). DINOv2: Learning robust visual features without supervision. Transactions on Machine Learning Research.

[29] Hj, D. (n.d.). Baselines-for-Industrial-Bin-Picking-BOP2025. https://github.com/demianhj/Baselines-for-Industrial-Bin-Picking-BOP2025

Downloads

Published

2026-10-23

Issue

Section

Articles