CVPR Findings2026
POS-ISP: Pipeline Optimization at the Sequence Level for Task-aware ISP
Jiyun Won, Heemin Yang, Woohyeok Kim, Jungseul Ok, Sunghyun Cho
Abstract
Recent work has explored optimizing image signal processing (ISP) pipelines for various tasks by composing predefined modules and adapting them to task-specific objectives. However, jointly optimizing module sequences and parameters remains challenging. Existing approaches rely on neural architecture search (NAS) or step-wise reinforcement learning (RL), but NAS suffers from a traininginference mismatch, while step-wise RL leads to unstable training and high computational overhead due to stagewise decision-making. We propose POS-ISP, a sequencelevel RL framework that formulates modular ISP optimization as a global sequence prediction problem. Our method predicts the entire module sequence and its parameters in a single forward pass and optimizes the pipeline using a terminal task reward, eliminating the need for intermediate supervision and redundant executions. Experiments across multiple downstream tasks show that POS-ISP improves task performance while reducing computational cost, highlighting sequence-level optimization as a stable and efficient paradigm for task-aware ISP. The project page is available at https://w1jyun.github.io/POS-ISP
Ask Lune
Know this paper in minutes, not an afternoon.
Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.
Lineage
- AdaptiveISP: Learning an Adaptive Image Signal Processor for Object Detection cites CycleISP: Real Image Restoration via Improved Data Synthesis
- AdaptiveISP: Learning an Adaptive Image Signal Processor for Object Detection cites ReconfigISP: Reconfigurable Camera Image Processing Pipeline
- Invertible Image Signal Processing cites CycleISP: Real Image Restoration via Improved Data Synthesis
- Task-Aware Image Signal Processor for Advanced Visual Perception cites AdaptiveISP: Learning an Adaptive Image Signal Processor for Object Detection
- Task-Aware Image Signal Processor for Advanced Visual Perception cites ReconfigISP: Reconfigurable Camera Image Processing Pipeline
- Task-Aware Image Signal Processor for Advanced Visual Perception cites Invertible Image Signal Processing
Builds on6
- ReconfigISP: Reconfigurable Camera Image Processing PipelineKe Yu, Zexian Li, Yue Peng, Chen Change Loy et al.ICCV 2021 · 46 citations
- AdaptiveISP: Learning an Adaptive Image Signal Processor for Object DetectionYujin Wang, Tianyi Xu, Zhang Fan, Tianfan Xue et al.NeurIPS 2024 · 38 citations
- Task-Aware Image Signal Processor for Advanced Visual PerceptionKai Chen, Jin Xiao, Leheng Zhang, Kexuan Shi et al.CVPR 2026 · 3 citations
- CycleISP: Real Image Restoration via Improved Data SynthesisSyed Waqas Zamir, Aditya Arora, Salman H. Khan, Munawar Hayat et al.CVPR 2020
- Invertible Image Signal ProcessingYazhou Xing, Zian Qian, Qifeng ChenCVPR 2021
Related papers
- RL-SeqISP: Reinforcement Learning-Based Sequential Optimization for Image Signal ProcessingXinyu Sun, Zhikun Zhao, Lili Wei, Congyan Lang et al.AAAI 2024 · 13 citations
- Learning to Exploit the Sequence-Specific Prior Knowledge for Image Processing Pipelines OptimizationHaina Qin, Longfei Han, Weihua Xiong, Juan Wang et al.CVPR 2023
- Multimodal Large Language Model-Guided ISP Hyperparameter Optimization with Dynamic Preference LearningXinyu Sun, Zhikun Zhao, Congyan Lang, Bing Li et al.ICCV 2025 · 1 citation
- DynamicISP: Dynamically Controlled Image Signal Processor for Image RecognitionMasakazu Yoshimura, Junji Otsuka, Atsushi Irie, Takeshi OhashiICCV 2023 · 28 citations
- MAS-ISP: A Proxy-Free Online Hyperparameter Optimization Framework for ISP Hardware SystemJiaming Liu, Xuan Huang, Zhijian Hao, Ruoxi Zhu et al.DAC 2025 · 3 citations
