Predicting High-precision Depth on Low-Precision Devices Using 2D Hilbert Curves

Mykhail Uss, Ruslan Yermolenko, Oleksii Shashko, Olena Kolodiazhna, Ivan Safonov, Volodymyr Savin, Yoonjae Yeo, Seowon Ji, Jaeyun Jeong
Proceedings of the 42nd International Conference on Machine Learning, PMLR 267:60635-60656, 2025.

Abstract

Dense depth prediction deep neural networks (DNN) have achieved impressive results for both monocular and binocular data but they are limited by high computational complexity, restricting their use on low-end devices. For better on-device efficiency and hardware utilization, weights and activations of the DNN should be converted to low-bit precision. However, this precision is not sufficient for representing high dynamic range depth. In this paper, we aim to overcome this limitation and restore high-precision depth from low-bit precision predictions. To achieve this, we propose to represent high dynamic range depth as two low dynamic range components of a Hilbert curve, and to train the full precision DNN to directly predict the latter. For on-device deployment, we use standard quantization methods and add a post-processing step that reconstructs depth from the Hilbert curve components predicted in low-bit precision. Extensive experiments demonstrate that our method increases bit precision of predicted depth by up to three bits with little computational overhead. We also observe a positive side effect of quantization error reduction by up to five times. Our method enables effective and accurate depth prediction with DNN weights and activations quantized to eight bit precision.

Cite this Paper

BibTeX
@InProceedings{pmlr-v267-uss25a, title = {Predicting High-precision Depth on Low-Precision Devices Using 2{D} {H}ilbert Curves}, author = {Uss, Mykhail and Yermolenko, Ruslan and Shashko, Oleksii and Kolodiazhna, Olena and Safonov, Ivan and Savin, Volodymyr and Yeo, Yoonjae and Ji, Seowon and Jeong, Jaeyun}, booktitle = {Proceedings of the 42nd International Conference on Machine Learning}, pages = {60635--60656}, year = {2025}, editor = {Singh, Aarti and Fazel, Maryam and Hsu, Daniel and Lacoste-Julien, Simon and Berkenkamp, Felix and Maharaj, Tegan and Wagstaff, Kiri and Zhu, Jerry}, volume = {267}, series = {Proceedings of Machine Learning Research}, month = {13--19 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v267/main/assets/uss25a/uss25a.pdf}, url = {https://proceedings.mlr.press/v267/uss25a.html}, abstract = {Dense depth prediction deep neural networks (DNN) have achieved impressive results for both monocular and binocular data but they are limited by high computational complexity, restricting their use on low-end devices. For better on-device efficiency and hardware utilization, weights and activations of the DNN should be converted to low-bit precision. However, this precision is not sufficient for representing high dynamic range depth. In this paper, we aim to overcome this limitation and restore high-precision depth from low-bit precision predictions. To achieve this, we propose to represent high dynamic range depth as two low dynamic range components of a Hilbert curve, and to train the full precision DNN to directly predict the latter. For on-device deployment, we use standard quantization methods and add a post-processing step that reconstructs depth from the Hilbert curve components predicted in low-bit precision. Extensive experiments demonstrate that our method increases bit precision of predicted depth by up to three bits with little computational overhead. We also observe a positive side effect of quantization error reduction by up to five times. Our method enables effective and accurate depth prediction with DNN weights and activations quantized to eight bit precision.} }
Endnote
%0 Conference Paper %T Predicting High-precision Depth on Low-Precision Devices Using 2D Hilbert Curves %A Mykhail Uss %A Ruslan Yermolenko %A Oleksii Shashko %A Olena Kolodiazhna %A Ivan Safonov %A Volodymyr Savin %A Yoonjae Yeo %A Seowon Ji %A Jaeyun Jeong %B Proceedings of the 42nd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2025 %E Aarti Singh %E Maryam Fazel %E Daniel Hsu %E Simon Lacoste-Julien %E Felix Berkenkamp %E Tegan Maharaj %E Kiri Wagstaff %E Jerry Zhu %F pmlr-v267-uss25a %I PMLR %P 60635--60656 %U https://proceedings.mlr.press/v267/uss25a.html %V 267 %X Dense depth prediction deep neural networks (DNN) have achieved impressive results for both monocular and binocular data but they are limited by high computational complexity, restricting their use on low-end devices. For better on-device efficiency and hardware utilization, weights and activations of the DNN should be converted to low-bit precision. However, this precision is not sufficient for representing high dynamic range depth. In this paper, we aim to overcome this limitation and restore high-precision depth from low-bit precision predictions. To achieve this, we propose to represent high dynamic range depth as two low dynamic range components of a Hilbert curve, and to train the full precision DNN to directly predict the latter. For on-device deployment, we use standard quantization methods and add a post-processing step that reconstructs depth from the Hilbert curve components predicted in low-bit precision. Extensive experiments demonstrate that our method increases bit precision of predicted depth by up to three bits with little computational overhead. We also observe a positive side effect of quantization error reduction by up to five times. Our method enables effective and accurate depth prediction with DNN weights and activations quantized to eight bit precision.
APA
Uss, M., Yermolenko, R., Shashko, O., Kolodiazhna, O., Safonov, I., Savin, V., Yeo, Y., Ji, S. & Jeong, J.. (2025). Predicting High-precision Depth on Low-Precision Devices Using 2D Hilbert Curves. Proceedings of the 42nd International Conference on Machine Learning, in Proceedings of Machine Learning Research 267:60635-60656 Available from https://proceedings.mlr.press/v267/uss25a.html.

Related Material