Research map / Label-efficient and robust learning / Self-supervised and contrastive
The Unreasonable Effectiveness of Deep Features as a Perceptual Metric
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 pretrained deep visual features instead of classic low-level metrics | BAPPS 2AFC perceptual similarity · BAPPS All | 2AFC accuracy: 63.8 -> 68.9 | Helped |
| Calibrate a pretrained representation with learned linear channel weights (ablation) | AlexNet · BAPPS Real Algorithms All | 2AFC accuracy: 65.0 -> 65.3 | Helped |
| Fine-tune all pretrained network weights instead of learning only linear channel weights (ablation) | AlexNet · BAPPS Real Algorithms All | 2AFC accuracy: 65.3 -> 64.3 | Hurt |
| Train AlexNet from scratch rather than use the pretrained AlexNet representation (ablation) | AlexNet · BAPPS Real Algorithms All | 2AFC accuracy: 65.0 -> 65.2 | No clear effect |
| Train AlexNet from scratch on perceptual judgments (ablation) | AlexNet · BAPPS distortions All | 2AFC accuracy: 76.8 -> 80.2 | Helped |
| Fine-tune all pretrained AlexNet weights on perceptual judgments (ablation) | AlexNet · BAPPS distortions All | 2AFC accuracy: 76.8 -> 80.6 | Helped |
| Train VGG from scratch on perceptual judgments (ablation) | VGG · BAPPS distortions All | 2AFC accuracy: 75.7 -> 80.8 | Helped |
| Fine-tune all pretrained VGG weights on perceptual judgments (ablation) | VGG · BAPPS distortions All | 2AFC accuracy: 75.7 -> 81.4 | Helped |
| Use self-supervised BiGAN features rather than classic low-level metrics | BAPPS perceptual similarity · BAPPS All | 2AFC accuracy: 63.8 -> 68.4 | Helped |
| Use randomly initialized Gaussian network features rather than classic low-level metrics | Random network · BAPPS All | 2AFC accuracy: 64.3 vs 63.8 | No clear effect |
Closest papers by meaning
- The Many Senses of Visual Similarity: A Text-Prompted Image Perceptual Metric (NeurIPS 2026 · similarity 0.88)
- The Unreasonable Effectiveness of Linear Prediction as a Perceptual Metric (ICLR 2024 · similarity 0.88)
- When does perceptual alignment benefit vision representations? (NeurIPS 2024 · similarity 0.88)
- LipSim: A Provably Robust Perceptual Similarity Metric (ICLR 2024 · similarity 0.87)
- Measuring Per-Unit Interpretability at Scale Without Humans (NeurIPS 2024 · similarity 0.87)
- Representational Similarity via Interpretable Visual Concepts (ICLR 2025 · similarity 0.86)
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.