Please use this identifier to cite or link to this item:
https://doi.org/10.21256/zhaw-20419| Publication type: | Conference paper |
| Type of review: | Peer review (publication) |
| Title: | Combining reinforcement learning with supervised deep learning for neural active scene understanding |
| Authors: | Roost, Dano Meier, Ralph Toffetti Carughi, Giovanni Stadelmann, Thilo |
| et. al: | No |
| DOI: | 10.21256/zhaw-20419 |
| Conference details: | Active Vision and Perception in Human(-Robot) Collaboration Workshop at IEEE RO-MAN 2020 (AVHRC’20), online, 31 August - 4 September 2020 |
| Issue Date: | 31-Aug-2020 |
| Publisher / Ed. Institution: | University of Essex |
| Language: | English |
| Subjects: | Active Vision; Deep Learning; Reinforcement Learning; Neural Scene Understanding; Robotic Grasping; Computer Vision |
| Subject (DDC): | 006: Special computer methods |
| Abstract: | While vision in living beings is an active process where image acquisition and classification are intertwined to gradually refine perception, much of today’s computer vision is build on the inferior paradigm of episodic classification of i.i.d. samples. We aim at improved scene understanding for robots by taking the sequential nature of seeing over time into account. We present a supervised multi-task approach to answer questions about different aspects of a scene such as the relationship between objects, their quantity or the their relative positions to the camera. For each question, we train a different output head which operates on input from one shared recurrent convolutional neural network that accumulates information over time steps. In parallel, we train an additional output head using reinforcement learning (RL) that uses the reduction in cumulative loss from the supervised heads as reward signal. It thereby learns to gradually improve the prediction confidence of e.g. partially occluded objects by moving the camera to a more favourable angle with respect to these objects. We present preliminary results on simulated RGB-D image sequences that show superior performance of our RL-based approach in answering questions quicker and more accurately than using static or random camera movement. |
| Further description: | Awarded with the Dr. Waldemar Jucker award 2020 of the GST |
| URI: | https://digitalcollection.zhaw.ch/handle/11475/20419 |
| Fulltext version: | Accepted version |
| License (according to publishing contract): | Licence according to publishing contract |
| Departement: | School of Engineering |
| Organisational Unit: | Institute of Computer Science (InIT) |
| Appears in collections: | Publikationen School of Engineering |
Files in This Item:
| File | Description | Size | Format | |
|---|---|---|---|---|
| 2020_Roost_Combining_reinforcement_learning_with_supervised_deep_learning.pdf | Accepted Version | 1.52 MB | Adobe PDF |  View/Open |
Show full item record
Roost, D., Meier, R., Toffetti Carughi, G., & Stadelmann, T. (2020, August 31). Combining reinforcement learning with supervised deep learning for neural active scene understanding. Active Vision and Perception in Human(-Robot) Collaboration Workshop at IEEE RO-MAN 2020 (AVHRC’20), Online, 31 August - 4 September 2020. https://doi.org/10.21256/zhaw-20419
Roost, D. et al. (2020) ‘Combining reinforcement learning with supervised deep learning for neural active scene understanding’, in Active Vision and Perception in Human(-Robot) Collaboration Workshop at IEEE RO-MAN 2020 (AVHRC’20), online, 31 August - 4 September 2020. University of Essex. Available at: https://doi.org/10.21256/zhaw-20419.
D. Roost, R. Meier, G. Toffetti Carughi, and T. Stadelmann, “Combining reinforcement learning with supervised deep learning for neural active scene understanding,” in Active Vision and Perception in Human(-Robot) Collaboration Workshop at IEEE RO-MAN 2020 (AVHRC’20), online, 31 August - 4 September 2020, Aug. 2020. doi: 10.21256/zhaw-20419.
ROOST, Dano, Ralph MEIER, Giovanni TOFFETTI CARUGHI und Thilo STADELMANN, 2020. Combining reinforcement learning with supervised deep learning for neural active scene understanding. In: Active Vision and Perception in Human(-Robot) Collaboration Workshop at IEEE RO-MAN 2020 (AVHRC’20), online, 31 August - 4 September 2020. Conference paper. University of Essex. 31 August 2020
Roost, Dano, Ralph Meier, Giovanni Toffetti Carughi, and Thilo Stadelmann. 2020. “Combining Reinforcement Learning with Supervised Deep Learning for Neural Active Scene Understanding.” Conference paper. In Active Vision and Perception in Human(-Robot) Collaboration Workshop at IEEE RO-MAN 2020 (AVHRC’20), Online, 31 August - 4 September 2020. University of Essex. https://doi.org/10.21256/zhaw-20419.
Roost, Dano, et al. “Combining Reinforcement Learning with Supervised Deep Learning for Neural Active Scene Understanding.” Active Vision and Perception in Human(-Robot) Collaboration Workshop at IEEE RO-MAN 2020 (AVHRC’20), Online, 31 August - 4 September 2020, University of Essex, 2020, https://doi.org/10.21256/zhaw-20419.
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.