Research map / Image generation / Editing and style transfer
Parameter Efficient Local Implicit Image Function Network for Face Segmentation
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 a local implicit-function decoder instead of a Conv U-Net type decoder (ablation) | FP-LIIF · LaPa | mean F1: 84.9 compared to 92.4 | Helped |
| use instance normalization in the modified residual blocks rather than the original EDSR block (ablation) | FP-LIIF · LaPa | mean F1: 92.3 -> 92.4 | Helped |
| use batch normalization in the modified residual blocks rather than the original EDSR block (ablation) | FP-LIIF · LaPa | mean F1: 92.3 -> 92.32 | Helped |
| add edge-aware cross-entropy loss (λ=40) rather than using λ=0 (ablation) | FP-LIIF · LaPa | mean F1: 91.73 -> 92.4 | Helped |
| increase edge-aware loss weight from λ=10 to λ=40 (ablation) | FP-LIIF · LaPa | mean F1: 92.2 -> 92.4 | Helped |
| generate 512-resolution output without changing the 256×256 input (ablation) | FP-LIIF · LaPa | mean F1: 92.41 -> 92.38 | No clear effect |
| generate 192-resolution output and upsample it to 256 (ablation) | FP-LIIF · LaPa | mean F1: 92.41 -> 92.35 | No clear effect |
| generate 128-resolution output and upsample it to 256 (ablation) | FP-LIIF · LaPa | mean F1: 92.41 -> 92.23 | Hurt |
| generate 96-resolution output and upsample it to 256 (ablation) | FP-LIIF · LaPa | mean F1: 92.41 -> 91.76 | Hurt |
| generate 64-resolution output and upsample it to 256 (ablation) | FP-LIIF · LaPa | mean F1: 92.41 -> 90.12 | Hurt |
| generate 512-resolution output without changing the 256×256 input (ablation) | FP-LIIF · CelebAMask-HQ | mean F1: 86.07 -> 86.14 | Helped |
| generate 192-resolution output and upsample it to 256 (ablation) | FP-LIIF · CelebAMask-HQ | mean F1: 86.07 -> 86.05 | No clear effect |
Closest papers by meaning
- Towards Learning Structure via Consensus for Face Segmentation and Parsing (CVPR 2020 · similarity 0.88)
- A New Dataset and Boundary-Attention Semantic Segmentation for Face Parsing (AAAI 2020 · similarity 0.87)
- SegFace: Face Segmentation of Long-Tail Classes (AAAI 2025 · similarity 0.87)
- Face Parsing With RoI Tanh-Warping (CVPR 2019 · similarity 0.86)
- Self-Supervised Facial Representation Learning with Facial Region Awareness (CVPR 2024 · similarity 0.85)
- MaSS13K: A Matting-level Semantic Segmentation Benchmark (CVPR 2025 · 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.