Elsa Lab

[Two CVPR 2024 Papers Accepted]

Two papers from Elsa Lab members and our collaborators have been accepted for presentation by CVPR 2024.

- Confronting Ambiguity in 6D Object Pose Estimation via Score-Based Diffusion on SE(3). Congratulations to 蕭謦, 陳澔威, 楊炫恭, and Chun-Yi Lee!

- Boosting Flow-based Generative Super-Resolution Models via Learned Prior. Congratulations to Li-Yuan Tsao, 陳澔威, 馮謙, Chun-Yi Lee, and our collaborators 羅以宸, 張嘉哲, Roy Tseng from MediaTek!

We appreciate all the supports from the NSTC, MediaTek, and NCHC for the computational resources.

[ICLR 2022] Denoising Likelihood Score Matching for Condition Score-Based Data Generation

We propose a new denoising likelihood score-matching (DLSM) loss to deal with the score mismatch issue we found in the existing conditional score-based data generation methods.
Advanced detail please visit: https://bit.ly/3MYLBkq
Paper Download: https://bit.ly/3N0At6p
arXiv: https://bit.ly/35mIPo8

ELSA Lab is a research laboratory focusing on Deep Reinforcement Learning, Intelligent Robotics, and Computer Vision. Please visit our website: https://elsalab.ai/

[ICLR 2022] Denoising Likelihood Score Matching for Condition Score-Based Data Generation ICLR 2022 Full Paper

[AAMAS 2018] A Deep Policy Inference Q-Network for Multi-Agent Systems
We present DPIQN, a deep policy inference Q-network that targets multi-agent systems composed of controllable agents, collaborators, and opponents that interact with each other.

Advanced detail please visit: https://bit.ly/3ugdw8l
Paper Download: https://bit.ly/34rifZZ

ELSA Lab is a research laboratory focusing on Deep Reinforcement Learning, Intelligent Robotics, and Computer Vision. Please visit our website: https://elsalab.ai/

[AAMAS 2018] A Deep Policy Inference Q-Network for Multi-Agent Systems AAMAS 2018 Full Paper

[CoRL 2019] Adversarial Active Exploration for Inverse Dynamics Model Learning
We presented an adversarial active exploration, which consists of a DRL agent and an inverse dynamics model competing with each other for efficient data collection.

Advanced detail please visit: https://bit.ly/3Hd0ITo
Paper Download: https://bit.ly/3yVHdMh

ELSA Lab is a research laboratory focusing on Deep Reinforcement Learning, Intelligent Robotics, and Computer Vision. Please visit our website: https://elsalab.ai/

[CoRL 2019] Adversarial Active Exploration for Inverse Dynamics Model Learning CoRL 2019 Full Paper

[NeurIPS 2018] Diversity-Driven Exploration Strategy for Deep Reinforcement Learning

​​We presented a diversity-driven exploration strategy, which can be effectively combined with current DRL algorithms through using an additional distance measure term to the loss function.

Advanced detail please visit: https://bit.ly/2ZmCLZE
Paper Download: https://bit.ly/3DT8Mrp
Demonstration Video: https://bit.ly/3nSLE6U

ELSA Lab is a research laboratory focusing on Deep Reinforcement Learning, Intelligent Robotics, and Computer Vision. Please visit our website: https://elsalab.ai/

[NeurIPS 2018] Diversity-Driven Exploration Strategy for Deep Reinforcement Learning NeurIPS 2018 Full Paper

[ICML 2021 Spotlight] DFAC Framework: Factorizing the Value Function via Quantile Mixture for Multi-Agent Distributional Q-Learning
We provided a distributional perspective on value function factorization methods, and introduced a framework, called DFAC, for integrating distributional RL with MARL domains. We achieve State-of-the-art performance on the 5 Super Hard scenarios in the SMAC benchmark.

Advanced detail please visit: https://bit.ly/2YBeDC8
Paper Download: https://reurl.cc/Xl4NR0
GitHub: https://reurl.cc/EZpL6k
Presentation Video: https://reurl.cc/82WL6j
Demonstration Video: https://reurl.cc/vge4pL

ELSA Lab is a research laboratory focusing on Deep Reinforcement Learning, Intelligent Robotics, and Computer Vision. Please visit our website: https://elsalab.ai/

[ICML 2021 Spotlight] DFAC Framework: Factorizing the Value Function via Quantile Mixture for… ICML 2021 Full Paper

❗️⏰❗️The deadline for submission is 15 October 2021. Both unpublished and already-published works are welcome! 🤖️

[MRVC-21] Workshop on Machine Learning for Mobile Robot MRVC workshop brings researchers in computer vision, machine learning, and robotics communities together to discuss the challenges and opportunities for mobile robots.

[Call for Submission] ACML Workshop on Machine Learning for Mobile Robot Vision and Control (MRVC)

The submission site is open!🔥🔥🔥

At MRVC-21, we will solicit contributions at the intersection of mobile robotics, machine learning, and computer vision.🤖

Please visit: https://mrvc-2021.net

[MRVC-21] Workshop on Machine Learning for Mobile Robot MRVC workshop brings researchers in computer vision, machine learning, and robotics communities together to discuss the challenges and opportunities for mobile robots.

[CVPR 2018] Dynamic Video Segmentation Network

We present our Dynamic Video Segmentation Network (DVSNet) for fast and efficient semantic video segmentation. DVSNet utilizes a decision network based on expected confidence score to make decisions and forwards different frame regions to a more accurate but slower segmentation path or a less accurate but faster spatial warping path. DVSNet is able to strike a balance between quality and efficiency for semantic video segmentation.

Advanced detail please visit: https://bit.ly/3FuxT4Y
Paper Download: https://reurl.cc/Nr3G86
arXiv: https://reurl.cc/xGd1b1
GitHub: https://reurl.cc/Gm7XYW
Demonstration Video: https://reurl.cc/dGReb8

ELSA Lab is a research laboratory focusing on Deep Reinforcement Learning, Intelligent Robotics, and Computer Vision. Please visit our website: https://elsalab.ai/

[CVPR 2018] Dynamic Video Segmentation Network 2018 CVPR Full Paper

[IJCAI 2018] Virtual-to-Real: Learning to Control in Visual Semantic Segmentation

We proposed to separate the model into a perception module and a control policy module, and introduced the concept of using semantic image segmentation as the meta state for relating these two modules in order to transfer policies learned in simulators to the real world.

Advanced detail please visit: https://bit.ly/2YwoedJ
ArXiv: https://reurl.cc/GmyG3p
IJCAI: https://reurl.cc/9rNbKj
Video Link: https://reurl.cc/bX8NQy

ELSA Lab is a research laboratory focusing on Deep Reinforcement Learning, Intelligent Robotics, and Computer Vision. Please visit our website: https://elsalab.ai/

[IJCAI 2018] Virtual-to-Real: Learning to Control in Visual Semantic Segmentation 2018 IJCAI Full Paper

[ICCD 2019] A Distributed Scheme for Accelerating Semantic

Video Segmentation on An Embedded Cluster
We present a framework which is in a master-slave hierarchy for performing semantic video segmentation tasks on an embedded cluster with increasing frame rates and small accuracy degradation.

Advanced detail please visit: https://bit.ly/3oRuasx
Download: https://reurl.cc/VEe1vN
IEEE: https://reurl.cc/a9zGjY
Presentation Link: https://reurl.cc/O0D43R

ELSA Lab is a research laboratory focusing on Deep Reinforcement Learning, Intelligent Robotics, and Computer Vision. Please visit our website: https://elsalab.ai/

[ICCD 2019] A Distributed Scheme for Accelerating Semantic Video Segmentation on An Embedded… 2019 ICCD Full Paper

[ICRA 2021] Reducing the Deployment-Time Inference Control Costs of Deep Reinforcement Learning Agents via an Asymmetric Architecture

We proposed a methodology of DRL framework for performing cost-aware control based on an asymmetric architecture. Our methodology uses a master policy to select a small sub-policy to act when conditions are acceptable while employing a large one when necessary.

Advanced detail please visit: https://bit.ly/3lqpY0I
Paper Download: https://reurl.cc/XWGmpa
Github Link: https://reurl.cc/2rYE3m
Presentation Link: https://reurl.cc/xGa6o5

ELSA Lab is a research laboratory focusing on Deep Reinforcement Learning, Intelligent Robotics, and Computer Vision. Please visit our website: https://elsalab.ai/

[ICRA 2021] Reducing the Deployment-Time Inference Control Costs of Deep Reinforcement Learning… 2021 ICRA Full Paper