Real-Time and Accurate Self-Supervised Monocular Depth Estimation on Mobile Device

Hong Cai, Fei Yin, Tushar Singhal, Sandeep Pendyam, Parham Noorzad, Yinhao Zhu, Khoi Nguyen, Janarbek Matai, Bharath Ramaswamy, Frank Mayer, Chirag Patel, Abhijit Khobare, Fatih Porikli
Proceedings of the NeurIPS 2021 Competitions and Demonstrations Track, PMLR 176:308-313, 2022.

Abstract

In this paper, we present our innovations on self-supervised monocular depth estimation. First, we enhance self-supervised monocular depth estimation with semantic information during training. This reduces the error by 12% and achieves state-of-the-art performance. Second, we enhance the backbone architecture using a scalable method for neural architecture search which optimizes directly for inference latency on a target device. This enables operation at more than 30 FPS. We demonstrate these techniques on a smartphone powered by a Snapdragon Mobile Platform.

Cite this Paper

BibTeX
@InProceedings{pmlr-v176-cai22a, title = {Real-Time and Accurate Self-Supervised Monocular Depth Estimation on Mobile Device}, author = {Cai, Hong and Yin, Fei and Singhal, Tushar and Pendyam, Sandeep and Noorzad, Parham and Zhu, Yinhao and Nguyen, Khoi and Matai, Janarbek and Ramaswamy, Bharath and Mayer, Frank and Patel, Chirag and Khobare, Abhijit and Porikli, Fatih}, booktitle = {Proceedings of the NeurIPS 2021 Competitions and Demonstrations Track}, pages = {308--313}, year = {2022}, editor = {Kiela, Douwe and Ciccone, Marco and Caputo, Barbara}, volume = {176}, series = {Proceedings of Machine Learning Research}, month = {06--14 Dec}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v176/cai22a/cai22a.pdf}, url = {https://proceedings.mlr.press/v176/cai22a.html}, abstract = {In this paper, we present our innovations on self-supervised monocular depth estimation. First, we enhance self-supervised monocular depth estimation with semantic information during training. This reduces the error by 12% and achieves state-of-the-art performance. Second, we enhance the backbone architecture using a scalable method for neural architecture search which optimizes directly for inference latency on a target device. This enables operation at more than 30 FPS. We demonstrate these techniques on a smartphone powered by a Snapdragon Mobile Platform.} }
Endnote
%0 Conference Paper %T Real-Time and Accurate Self-Supervised Monocular Depth Estimation on Mobile Device %A Hong Cai %A Fei Yin %A Tushar Singhal %A Sandeep Pendyam %A Parham Noorzad %A Yinhao Zhu %A Khoi Nguyen %A Janarbek Matai %A Bharath Ramaswamy %A Frank Mayer %A Chirag Patel %A Abhijit Khobare %A Fatih Porikli %B Proceedings of the NeurIPS 2021 Competitions and Demonstrations Track %C Proceedings of Machine Learning Research %D 2022 %E Douwe Kiela %E Marco Ciccone %E Barbara Caputo %F pmlr-v176-cai22a %I PMLR %P 308--313 %U https://proceedings.mlr.press/v176/cai22a.html %V 176 %X In this paper, we present our innovations on self-supervised monocular depth estimation. First, we enhance self-supervised monocular depth estimation with semantic information during training. This reduces the error by 12% and achieves state-of-the-art performance. Second, we enhance the backbone architecture using a scalable method for neural architecture search which optimizes directly for inference latency on a target device. This enables operation at more than 30 FPS. We demonstrate these techniques on a smartphone powered by a Snapdragon Mobile Platform.
APA
Cai, H., Yin, F., Singhal, T., Pendyam, S., Noorzad, P., Zhu, Y., Nguyen, K., Matai, J., Ramaswamy, B., Mayer, F., Patel, C., Khobare, A. & Porikli, F.. (2022). Real-Time and Accurate Self-Supervised Monocular Depth Estimation on Mobile Device. Proceedings of the NeurIPS 2021 Competitions and Demonstrations Track, in Proceedings of Machine Learning Research 176:308-313 Available from https://proceedings.mlr.press/v176/cai22a.html.

Related Material