Research map / Detection and segmentation / CNN features
Training Deep Networks to Be Spatially Sensitive
Nicholas Kolkin, Eli Shechtman, Gregory Shakhnarovich
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 cross-entropy with approximate weighted F-measure loss (ablation) | SZN · MSRA-B | F^w_1: 0.829 -> 0.856 | Improved |
| replace cross-entropy with approximate weighted F-measure loss (ablation) | SZN · HKU-IS | F^w_1: 0.788 -> 0.826 | Improved |
| replace cross-entropy with approximate weighted F-measure loss (ablation) | SZN · ECSSD | F^w_1: 0.796 -> 0.827 | Improved |
| replace cross-entropy with approximate weighted F-measure loss (ablation) | SZN · PASCAL-S | F^w_1: 0.657 -> 0.680 | Improved |
| replace cross-entropy with approximate weighted F-measure loss (ablation) | SZN · ECSSD | MAE: 0.069 -> 0.073 | No clear effect |
| replace cross-entropy with approximate weighted F-measure loss (ablation) | SZN · Portrait segmentation | MIoU: 96.53 -> 97.13 | Improved |
| add the foreground reweighting matrix A (ablation) | Proposed loss · MSRA-B objective comparison | F^w_1: 0.836 -> 0.856 | Improved |
| add the false-positive boundary weighting B (ablation) | Proposed loss · MSRA-B objective comparison | F^w_1: 0.835 -> 0.856 | Improved |
| use proposed loss instead of CE with 2x foreground weight (ablation) | SZN · MSRA-B objective comparison | AUROC: 0.976 -> 0.988 | Improved |
| use proposed loss instead of CE ignoring 3-pixel border band (ablation) | SZN · MSRA-B objective comparison | F^w_1: 0.807 -> 0.856 | Improved |
| use proposed loss instead of CE with 2x background weight (ablation) | SZN · MSRA-B objective comparison | F^w_1: 0.797 -> 0.856 | Improved |
| approximate F^w_beta with convolutions | F^w_beta metric · 15 ECSSD images | computation time: 37 minutes -> 0.33 seconds | Improved |
Closest papers by meaning
- Structure-Measure: A New Way to Evaluate Foreground Maps (ICCV 2017 · similarity 0.86)
- Optimizing the F-Measure for Threshold-Free Salient Object Detection (ICCV 2019 · similarity 0.84)
- Role of Locality and Weight Sharing in Image-Based Tasks: A Sample Complexity Separation between CNNs, LCNs, and FCNs (ICLR 2024 · similarity 0.84)
- The Unreasonable Effectiveness of Linear Prediction as a Perceptual Metric (ICLR 2024 · similarity 0.84)
- Do Vision-Language Models Measure Up? Benchmarking Visual Measurement Reading with MeasureBench (CVPR 2026 · similarity 0.83)
- Measuring Per-Unit Interpretability at Scale Without Humans (NeurIPS 2024 · similarity 0.82)
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.
