Research map / Detection and segmentation / CNN features
Understanding Low- and High-Level Contributions to Fixation Prediction
Matthias Kummerer, Thomas S. A. Wallis, Leon A. Gatys, Matthias Bethge
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 |
|---|---|---|---|
| replace low-level isotropic contrast features with pre-trained VGG-19 object features under the same readout | ICF readout · MIT1003 (cross-validated) | information gain explained (%): 37.2 -> 80.3 | Improved |
| use ICF intensity-contrast features instead of hand-tuned or from-scratch models | Saliency model · MIT1003 (cross-validated) | information gain explained (%): 31.1 -> 37.2 | Improved |
| use pre-trained deep features (DeepGaze II) instead of earlier DeepGaze I | DeepGaze · MIT1003 (cross-validated) | information gain explained (%): 46.1 -> 80.3 | Improved |
| use DeepGaze II instead of SALICON on the held-out benchmark | Saliency model · MIT300 benchmark (held-out) | sAUC (%): 74% -> 77% | Improved |
| use DeepGaze II instead of DSCLRCN on NSS | Saliency model · MIT300 benchmark (held-out) | NSS: 2.35 -> 2.34 | Hurt |
| compare ICF to DeepGaze II per image | DeepGaze II · MIT1003 (cross-validated) | images better predicted: 94 of 1003 | Mixed |
| assess fixations individually: ICF vs DeepGaze II or center bias | DeepGaze II · MIT1003 (cross-validated) | share of fixations better explained: around 25% | Mixed |
Closest papers by meaning
- Inferring Salient Objects from Human Fixations (TPAMI 2019 · similarity 0.86)
- Saliency Unified: A Deep Architecture for Simultaneous Eye Fixation Prediction and Salient Object Segmentation (CVPR 2016 · similarity 0.86)
- Harmonizing the object recognition strategies of deep neural networks with humans (NeurIPS 2022 · similarity 0.85)
- Salient Object Detection Driven by Fixation Prediction (CVPR 2018 · similarity 0.85)
- No Pains, More Gains: Recycling Sub-Salient Patches for Efficient High-Resolution Image Recognition (CVPR 2025 · similarity 0.83)
- Interpreting Object-level Foundation Models via Visual Precision Search (CVPR 2025 · 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.
