Research map / Detection and segmentation / Weakly supervised detection and segmentation
Real Time Image Saliency for Black Box Classifiers
Piotr Dabkowski, Yarin Gal
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 a trained masking model to predict saliency maps in one forward pass | ImageNet localisation with GoogLeNet · ImageNet validation | localisation error (%): 43.1 -> 36.9 | Improved |
| Use a trained masking model rather than gradient saliency | ImageNet localisation with GoogLeNet · ImageNet validation | localisation error (%): 41.7 -> 36.9 | Improved |
| Use a trained masking model rather than Excitation Backprop | ImageNet localisation with GoogLeNet · ImageNet validation | localisation error (%): 39.0 -> 36.9 | Improved |
| Use a trained masking model rather than the Center box baseline | ImageNet localisation with GoogLeNet · ImageNet validation | localisation error (%): 46.3 -> 36.9 | Improved |
| Use a trained masking model rather than the Excitation Backprop map | ImageNet saliency evaluation with GoogLeNet · ImageNet validation | saliency metric: 0.415 -> 0.318 | Improved |
| Use a trained masking model rather than gradient saliency | ImageNet saliency evaluation with GoogLeNet · ImageNet validation | saliency metric: 0.451 -> 0.318 | Improved |
| Train the masking model on ResNet-50 rather than AlexNet | Masking model · ImageNet validation | localisation error (%): 39.8 -> 36.7 | Improved |
| Train the masking model on GoogLeNet rather than AlexNet | Masking model · ImageNet validation | localisation error (%): 39.8 -> 36.9 | Improved |
| Add random fake-class selectors during training and apply only the area penalty for them | Masking model training · ImageNet | class-selector conditioning | Improved |
| Randomly vary the evidence-removal image between blurred input and random noisy colour | Masking model training · ImageNet | mask quality | Improved |
| Use the two-channel absolute-ratio mask output instead of sigmoid or tanh (ablation) | Masking model · ImageNet | training optimisation | Improved |
| Add total-variation smoothness regularisation to the mask objective | Masking model · ImageNet saliency maps | artifact suppression | Improved |
Closest papers by meaning
- Non-Local Deep Features for Salient Object Detection (CVPR 2017 · similarity 0.89)
- Deep Saliency With Encoded Low Level Distance Map and High Level Features (CVPR 2016 · similarity 0.88)
- ELD-Net: An Efficient Deep Learning Architecture for Accurate Saliency Detection (TPAMI 2017 · similarity 0.88)
- Salient Object Detection: A Discriminative Regional Feature Integration Approach (IJCV 2016 · similarity 0.88)
- Learning to Promote Saliency Detectors (CVPR 2018 · similarity 0.88)
- Salient Object Detection Driven by Fixation Prediction (CVPR 2018 · similarity 0.87)
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.
