Research map / Detection and segmentation / CNN features
Attention-Aware Learning for Hyperparameter Prediction in Image Processing Pipelines
Haina Qin, Longfei Han, Juan Wang, Congxuan Zhang, Yanwei Li, Bing Li, Weiming Hu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use diverse per-image predicted parameters instead of fixed ones (ablation) | Attention-aware ISP predictor · Object detection (COCO) | mAP0.5: 0.57 -> 0.61 | Improved |
| use diverse per-image predicted parameters instead of fixed ones (ablation) | Attention-aware ISP predictor · Instance segmentation (COCO) | mAP0.5: 0.48 -> 0.52 | Improved |
| add multi-attention path (ablation) | Parameter prediction network · Object detection (COCO) | mAP0.5: 0.57 -> 0.61 | Improved |
| add multi-attention path (ablation) | Parameter prediction network · Instance segmentation (COCO) | mAP0.5: 0.47 -> 0.52 | Improved |
| use per-parameter multi-attention instead of single shared attention (ablation) | Parameter prediction network · Object detection (COCO) | mAP0.5: 0.58 -> 0.61 | Improved |
| use per-parameter multi-attention instead of single shared attention (ablation) | Parameter prediction network · Instance segmentation (COCO) | mAP0.5: 0.50 -> 0.52 | Improved |
| use multi-attention instead of self-attention (ablation) | Parameter prediction network · Object detection (COCO) | mAP0.5: 0.57 -> 0.61 | Improved |
| use multi-attention instead of self-attention (ablation) | Parameter prediction network · Instance segmentation (COCO) | mAP0.5: 0.48 -> 0.52 | Improved |
| use predicted per-image parameters instead of expert-tuned | Synthetic ISP · Object detection (COCO) | mAP0.5: 0.56 -> 0.61 | Improved |
| use predicted per-image parameters instead of expert-tuned | Synthetic ISP · Instance segmentation (COCO) | mAP0.5: 0.46 -> 0.52 | Improved |
| use predicted per-image parameters instead of expert-tuned | Synthetic ISP · Human viewing PSNR | PSNR (dB): 17.11 -> 19.96 | Improved |
Closest papers by meaning
- Learning To Exploit the Sequence-Specific Prior Knowledge for Image Processing Pipelines Optimization (CVPR 2023 · similarity 0.93)
- Multimodal Large Language Model-Guided ISP Hyperparameter Optimization with Dynamic Preference Learning (ICCV 2025 · similarity 0.87)
- RL-SeqISP: Reinforcement Learning-Based Sequential Optimization for Image Signal Processing (AAAI 2024 · similarity 0.86)
- AdaptiveISP: Learning an Adaptive Image Signal Processor for Object Detection (NeurIPS 2024 · similarity 0.85)
- Skip-Attention: Improving Vision Transformers by Paying Less Attention (ICLR 2024 · similarity 0.84)
- Principled Architecture-aware Scaling of Hyperparameters (ICLR 2024 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
