Research map / Detection and segmentation / CNN features
Personalized Saliency and Its Prediction
Yanyu Xu, Shenghua Gao, Junru Wu, Nianyi Li, Jingyi Yu A
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 |
|---|---|---|---|
| predict PSM as USM plus discrepancy (CNN-PIEF) instead of directly from RGB | CNN-PIEF · PSM dataset, closed-set | CC: 0.6109 -> 0.6532 | Improved |
| use discrepancy-based CNN-PIEF instead of per-observer MultiConvNets (ML-Net) | ML-Net · PSM dataset, closed-set | CC: 0.4310 -> 0.6368 | Improved |
| use shared Multi-task CNN instead of MultiConvNets (BMS) | BMS · PSM dataset, closed-set | CC: 0.5510 -> 0.6390 | Improved |
| use CNN-PIEF instead of MultiConvNets (SalNet) | SalNet · PSM dataset, closed-set | CC: 0.6397 -> 0.6771 | Improved |
| jointly predict USM and discrepancy (Baseline 1) instead of feeding predicted USM | RGB-only baseline · PSM dataset, closed-set | CC: 0.6109 -> 0.5703 | Hurt |
| replace predicted USM with ground-truth USM | CNN-PIEF (SalNet) · PSM dataset, closed-set | CC: 0.6771 -> 0.8651 | Improved |
| fine-tune SalNet on the 1000-image PSM training set | SalNet · PSM dataset, closed-set | CC: 0.6238 -> 0.5060 | Hurt |
| add supervision on middle layers (ablation) | Multi-task CNN and CNN-PIEF · PSM dataset, closed-set | CC | Improved |
| insert person-specific filters at conv5 instead of conv7 (ablation) | LDS based CNN-PIEF · PSM dataset, closed-set (insert position) | CC: 0.6163 -> 0.6532 | Improved |
| insert person-specific filters at all of conv5-7 instead of conv5 only (ablation) | LDS based CNN-PIEF · PSM dataset, closed-set (insert position) | CC: 0.6532 -> 0.5867 | Hurt |
| train on 20 observers and transfer to unseen observers instead of training on the open-set | SalNet based CNN-PIEF · PSM dataset, open-set (10 unseen observers) | CC: 0.6863 -> 0.6468 | Hurt |
| use different methods for USM and discrepancy prediction (LDS USM, ML-Net discrepancy) (ablation) | Multi-task CNN · PSM dataset, closed-set | CC: 0.6463 -> 0.5351 | Hurt |
Closest papers by meaning
- Learning From Unique Perspectives: User-Aware Saliency Modeling (CVPR 2023 · similarity 0.92)
- Saliency Prediction in the Deep Learning Era: Successes and Limitations (TPAMI 2019 · similarity 0.91)
- What Do Deep Saliency Models Learn about Visual Attention? (NeurIPS 2023 · similarity 0.91)
- A Deeper Look at Saliency: Feature Contrast, Semantics, and Beyond (CVPR 2016 · similarity 0.90)
- Understanding and Visualizing Deep Visual Saliency Models (CVPR 2019 · similarity 0.90)
- CaRDiff: Video Salient Object Ranking Chain of Thought Reasoning for Saliency Prediction with Diffusion (AAAI 2025 · similarity 0.88)
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.
