Papers
Topics
Authors
Recent
Search
2000 character limit reached

Modular Quantization-Aware Training for 6D Object Pose Estimation

Published 12 Mar 2023 in cs.CV, cs.LG, and cs.RO | (2303.06753v3)

Abstract: Edge applications, such as collaborative robotics and spacecraft rendezvous, demand efficient 6D object pose estimation on resource-constrained embedded platforms. Existing 6D pose estimation networks are often too large for such deployments, necessitating compression while maintaining reliable performance. To address this challenge, we introduce Modular Quantization-Aware Training (MQAT), an adaptive and mixed-precision quantization-aware training strategy that exploits the modular structure of modern 6D pose estimation architectures. MQAT guides a systematic gradated modular quantization sequence and determines module-specific bit precisions, leading to quantized models that outperform those produced by state-of-the-art uniform and mixed-precision quantization techniques. Our experiments showcase the generality of MQAT across datasets, architectures, and quantization algorithms. Remarkably, MQAT-trained quantized models achieve a significant accuracy boost (>7%) over the baseline full-precision network while reducing model size by a factor of 4x or more. Our project website is at: https://saqibjaved1.github.io/MQAT_/

Definition Search Book Streamline Icon: https://streamlinehq.com
References (59)
  1. Composable sparse fine-tuning for cross-lingual transfer. arXiv preprint arXiv:2110.07560, 2021.
  2. Lsq+: Improving low-bit quantization through learnable offsets and better initialization. Computer Vision and Pattern Recognition, 2020.
  3. What is the state of neural network pruning? Machine Learning and Systems, 2020.
  4. Learning 6d object pose estimation using 3d object coordinates. European Conference on Computer Vision, 2014.
  5. Zeroq: A novel zero shot quantization framework. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 13169–13178, 2020.
  6. Rethinking differentiable search for mixed-precision neural networks. In Computer Vision and Pattern Recognition, 2020.
  7. Satellite pose estimation with deep landmark regression and nonlinear pose refinement. IEEE International Conference on Computer Vision Workshop, pages 2816–2824, 2019.
  8. Aqd: Towards accurate quantized object detection. In Computer Vision and Pattern Recognition, 2021a.
  9. Towards mixed-precision quantization of neural networks via constrained optimization, 2021b.
  10. Differentiable model compression via pseudo quantization noise. arXiv preprint arXiv:2104.09987, 2021.
  11. So-pose: Exploiting self-occulsion for direct 6d pose estimation. International Conference on Computer Vision, 2021.
  12. Hawq: Hessian aware quantization of neural networks with mixed-precision. IEEE International Conference on Computer Vision, 2019.
  13. Hawq-v2: Hessian aware trace-weighted quantization of neural networks. In Neural Information Processing Systems, 2020.
  14. Learned step size quantization. International Conference on Learning Representations, 2020.
  15. Optimal brain compression: A framework for accurate post-training quantization and pruning, 2022.
  16. Model based training, detection and pose estimation of texture-less 3d objects in heavily cluttered scenes. Asian Conference on Computer Vision, 2012.
  17. EPOS: Estimating 6D pose of objects with symmetries. IEEE Conference on Computer Vision and Pattern Recognition, 2020.
  18. On evaluation of 6d object pose estimation. European Conference on Computer Vision.
  19. Lora: Low-rank adaptation of large language models. arXiv preprint arXiv:2106.09685, 2021a.
  20. Segmentation-driven 6d object pose estimation. 2019.
  21. Wide-depth-range 6d object pose estimation in space. In Computer Vision and Pattern Recognition, 2021b.
  22. Repose: Fast 6d object pose refinement via deep texture rendering. In IEEE International Conference on Computer Vision, pages 3303–3312, 2021.
  23. Adaptive mixtures of local experts. Neural computation, 3(1):79–87, 1991.
  24. ipose: Instance-aware 6d pose estimation of partly occluded objects. In ACCV, 2018.
  25. Ssd-6d: Making rgb-based 3d detection and 6d pose estimation great again. 2017.
  26. Cosypose: Consistent multi-view multi-object 6d pose estimation. 2020.
  27. Brecq: Pushing the limit of post-training quantization by block reconstruction. International Conference on Learning Representations, 2021.
  28. Cdpn: Coordinates-based disentangled pose network for real-time rgb-based 6-dof object pose estimation. 2019.
  29. Feature pyramid networks for object detection. In Computer Vision and Pattern Recognition, 2016.
  30. Ssd: Single shot multibox detector. In European Conference on Computer Vision, 2016.
  31. Data-free quantization through weight equalization and bias correction. IEEE International Conference on Computer Vision, pages 1325–1334, 2019.
  32. Up or down? adaptive rounding for post-training quantization. International Conference on Machine Learning, 2020.
  33. Making deep heatmaps robust to partial occlusions for 3d object pose estimation. In Proc. of European Conference on Computer Vision, 2018.
  34. Pvnet: Pixel-wise voting network for 6dof pose estimation. Computer Vision and Pattern Recognition, 2019.
  35. Robot guidance using machine vision techniques in industrial environmnets: A comparative review. Sensors, 16(3):335, 2016.
  36. Qbitopt: Fast and accurate bitwidth reallocation during training. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 1282–1291, 2023.
  37. Mad-x: An adapter-based framework for multi-task cross-lingual transfer. arXiv preprint arXiv:2005.00052, 2020.
  38. Modular deep learning. arXiv preprint arXiv:2302.11529, 2023.
  39. Mobile-ursonet: an embeddable neural network for onboard spacecraft pose estimation. arXiv preprint arXiv:2205.02065, 2022.
  40. Deep learning for spacecraft pose estimation from photorealistic rendering. International Conference on Robotics and Automation, 2020.
  41. Bb8: A scalable, accurate, robust to partial occlusion method for predicting the 3d poses of challenging objects without using depth. 2017.
  42. Nipq: Noise proxy-based integrated pseudo-quantization. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 3852–3861, 2023.
  43. A survey on vision guided robotic systems with intelligent control strategies for autonomous tasks. Cogent Engineering, 9(1):1–44, 2022.
  44. Hybridpose: 6d object pose estimation under hybrid representations, 2020.
  45. Deep learning-based spacecraft relative navigation methods: A survey. Acta Astronautica, 191:22–40, 2022.
  46. Zebrapose: Coarse to fine surface encoding for 6dof object pose estimation. Computer Vision and Pattern Recognition, 2022.
  47. Mixed-precision neural network quantization via learned layer-wise importance. arXiv preprint arXiv:2203.08368, 2022.
  48. Real-Time Seamless Single Shot 6D Object Pose Prediction. In IEEE Conference on Computer Vision and Pattern Recognition, 2018.
  49. Pyrapose: Feature pyramids for fast and accurate object pose estimation under domain shift. In International Conference on Robotics and Automation, 2021.
  50. F. Vicentini. Collaborative robotics: A survey. https://doi.org/10.1115/1.4046238, 2021.
  51. Gdr-net: Geometry-guided direct regression networks for monocular 6d pose estimation. Computer Vision and Pattern Recognition, 2021.
  52. Ca-spacenet: Counterfactual analysis for 6d pose estimation in space. 2022 IEEE/RSJ International Conference on Intelligent Robots and Systems, 2022.
  53. Posecnn: A convolutional neural network for 6d object pose estimation in cluttered scenes. Robotics: Science and Systems, 2018.
  54. Kohei Yamamoto. Learnable companding quantization for accurate low-bit neural networks. Computer Vision and Pattern Recognition, 2021.
  55. Hawqv3: Dyadic neural network quantization. 2020.
  56. Dpod: 6d pose object detector and refiner. IEEE International Conference on Computer Vision, 2019.
  57. Qd-bev: Quantization-aware view-guided distillation for multi-view 3d object detection. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 3825–3835, 2023.
  58. Improving Neural Network Quantization without Retraining using Outlier Channel Splitting. International Conference on Machine Learning (ICML), pages 7543–7552, 2019.
  59. Incremental network quantization: Towards lossless cnns with low-precision weights. International Conference on Learning Representations, 2017.

Summary

No one has generated a summary of this paper yet.

Sign Up to Summarize

Paper to Video (Beta)

No one has generated a video about this paper yet.

Sign Up to Generate All Videos Subscribe on YouTube

Whiteboard

No one has generated a whiteboard explanation for this paper yet.

Sign Up to Generate

Open Problems

We haven't generated a list of open problems mentioned in this paper yet.

Generate Now

Continue Learning

We haven't generated follow-up questions for this paper yet.

Generate Now

Collections

Sign up for free to add this paper to one or more collections.

Sign Up