RL-SeqISP: Reinforcement Learning-Based Sequential Optimization for Image Signal Processing
Authors: Xinyu Sun, Zhikun Zhao, Lili Wei, Congyan Lang, Mingxuan Cai, Longfei Han, Juan Wang, Bing Li, Yuxuan Guo
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | These merits of the RL-Seq ISP model as well as its high efficiency are substantiated by comprehensive experiments on a wide range of downstream tasks, including two visual analysis tasks (instance segmentation and object detection), and image quality assessment (IQA), as compared with representative methods both quantitatively and qualitatively. |
| Researcher Affiliation | Collaboration | Xinyu Sun1,2*, Zhikun Zhao1,2*, Lili Wei1*, Congyan Long1 , Mingxuan Cai3, Longfei Han4, Juan Wang2, Bing Li2,5, Yuxuan Guo6 1Key Laboratory of Big Data & Artificial Intelligence in Transportation (Ministry of Education), School of Computer and Information Technology, Beijing Jiaotong University 2Institute of Automation, Chinese Academy of Sciences 3Shanghai Jiao Tong University 4Beijing Technology and Business University 5People AI Inc. Beijing, China 6 Shenzhen Heytap Technology Corp., Ltd |
| Pseudocode | Yes | Algorithm 1: Training procedure of the RL-Seq ISP |
| Open Source Code | No | The paper does not contain an explicit statement or a link indicating that the source code for the described methodology is publicly available. |
| Open Datasets | Yes | For instance segmentation and object detection, due to the lack of a large-scale RAW-RGB image dataset, we employ a simulation method (Kim et al. 2012) to modify MSCOCO dataset (Lin et al. 2014), which is also been adopted by other comparable sota methods. |
| Dataset Splits | Yes | We evaluate the model on the MSCOCO validation set and report the m AP score. |
| Hardware Specification | No | The paper mentions using a 'SONY IMX766 CMOS sensor to collect 252 RAW images', but this refers to data collection, not the specific hardware (like GPUs or CPUs) used for training or inference in the experiments. |
| Software Dependencies | No | The paper mentions using Mask-RCNN, YOLOv3, and the PPO algorithm, but it does not specify version numbers for these software components or any other libraries or programming languages. |
| Experiment Setup | Yes | where ϵ denote a hyperparameter (set to 0.2 following (Schulman et al. 2017)), ˆAt indicate an advantage function (i.e., ˆAt = (Rt Vt)), and rt(θπ) is an objective function to constrain on the size of the actor update, respectively. ... T is set to 8, 4 and 4 for instance segmentation, object detection and IQA, respectively. |