Research map / Detection and segmentation / CNN features
Generic Perceptual Loss for Modeling Structured Output Dependencies
Check 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 |
|---|---|---|---|
| Add generic perceptual loss | DeepLabV3+ with ResNet50 · Cityscapes, ADE20K, and Pascal VOC | mIoU: 80.09 % -> 80.70%; 42.72% -> 42.95%; 75.93% -> 76.79% | Helped |
| Add generic perceptual loss | WCE depth estimator with ResNet50 · NYUDV2 | Rel. (%): 14.5 -> 14.0 | Helped |
| Add generic perceptual loss on top of WCE + VNL | ResNet50 depth estimator · NYUDV2 | Rel. (%): 13.6 -> 13.2 | Helped |
| Add generic perceptual loss | CondInst with ResNet50 · MS COCO | Box AP: 36.91 -> 37.44 | Helped |
| Add generic perceptual loss | CondInst with ResNet50 · MS COCO | Mask AP: 33.42 -> 33.69 | Helped |
| Use pretrained VGG19 weights instead of random weights (ablation) | Perceptual loss for PSPNet ResNet18 · Cityscapes val | mIoU: 71.25 -> 71.19 | No clear effect |
| Use VGG19 rather than VGG11 as the perceptual network (ablation) | Perceptual loss for PSPNet ResNet18 · Cityscapes val | mIoU: 70.18 ± 0.11 -> 71.25 ± 0.04 | Helped |
| Increase the perceptual-network kernel size from 1 to 7 (ablation) | Five-layer perceptual network · Cityscapes val | mIoU: 70.71 ± 0.020 -> 71.16 ± 0.122 | Helped |
| Add five equally weighted block-level losses instead of only the final-layer loss (ablation) | Perceptual loss for PSPNet ResNet18 · Cityscapes val | mIoU: 70.88% -> 70.73% | Hurt |
| Use weighted multi-level losses instead of only the final-layer loss (ablation) | Perceptual loss for PSPNet ResNet18 · Cityscapes val | mIoU: 70.88% -> 70.93% | Helped |
| Use the proposed initialization rather than Xavier-normal (ablation) | Perceptual loss network · Cityscapes val | mIoU: 69.8 -> 71.3 | Helped |
| Use Gaussian initialization with standard deviation 1 (ablation) | Perceptual loss network · Cityscapes val | mIoU | Hurt |
Closest papers by meaning
- The Unreasonable Effectiveness of Deep Features as a Perceptual Metric (CVPR 2018 · similarity 0.87)
- Understanding and Simplifying Perceptual Distances (CVPR 2021 · similarity 0.86)
- Structured Convolutions for Efficient Neural Network Design (NeurIPS 2020 · similarity 0.86)
- Auto-Encoded Supervision for Perceptual Image Super-Resolution (CVPR 2025 · similarity 0.84)
- The Unreasonable Effectiveness of Linear Prediction as a Perceptual Metric (ICLR 2024 · similarity 0.84)
- Investigating the Benefits of Projection Head for Representation Learning (ICLR 2024 · similarity 0.84)
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.