Research map / Label-efficient and robust learning / Recognition and long-tail
DeepGaze IIE: Calibrated Prediction in and Out-of-Domain for State-of-the-Art Saliency Modeling
Akis Linardos, Matthias Kümmerer, Ori Press, 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 VGG19 backbone with ResNet50 features | DeepGaze II readout · MIT1003 | information gain (% of gold standard): 78% to 85% | Improved |
| use stronger ImageNet backbones such as EfficientNetB5 beyond ResNet50 | DeepGaze readout · MIT1003 | information gain | No clear effect |
| add DenseNet-201 to ResNext-50 mixture (ablation) | backbone mixture · MIT1003 | information gain: 1.0368 -> 1.1075 | Improved |
| add EfficientNet-B5 to ResNext-50 + DenseNet-201 (ablation) | backbone mixture · MIT1003 | information gain: 1.1075 -> 1.1256 | Improved |
| add ShapeNet-C to ResNext-50 + DenseNet-201 + EfficientNet-B5 (ablation) | backbone mixture · MIT1003 | information gain: 1.1256 -> 1.1285 | Improved |
| add ResNet-50 as fifth backbone to mixture (ablation) | DSRE · MIT1003 | information gain | Hurt |
| increase instances per backbone from 1 to 3 (ablation) | DSRE · MIT1003 | information gain: 1.1285 -> 1.13294 | Improved |
| increase instances per backbone from 3 to 5 (ablation) | DSRE · MIT1003 | information gain: 1.13294 -> 1.13287 | No clear effect |
| ensemble DeepGaze IIE vs DeepGaze II | saliency model · MIT300 | information gain: 0.9247 -> 1.0715 | Improved |
| use 4 instances per backbone ensemble instead of single instance on MIT300 (ablation) | DeepGaze IIE · MIT300 | information gain: 1.0679 -> 1.0715 | Improved |
| average only models with the same backbone (ablation) | ensemble · PASCAL-S/Toronto | calibration | No clear effect |
| fine-tune the backbone (ablation) | DeepGaze · MIT1003 | performance | Hurt |
Closest papers by meaning
- Modeling Saliency Dataset Bias (ICCV 2025 · similarity 0.88)
- Saliency Prediction in the Deep Learning Era: Successes and Limitations (TPAMI 2019 · similarity 0.88)
- Understanding and Visualizing Deep Visual Saliency Models (CVPR 2019 · similarity 0.86)
- Saliency Benchmarking Made Easy: Separating Models, Maps and Metrics (ECCV 2018 · similarity 0.86)
- Shallow and Deep Convolutional Networks for Saliency Prediction (CVPR 2016 · similarity 0.86)
- There and Back Again: Revisiting Backpropagation Saliency Methods (CVPR 2020 · similarity 0.85)
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.
