Research map / Image generation / Autoregressive and latent generation
Unveiling The Mask of Position-Information Pattern Through the Mist of Image Features
Chieh Hubert Lin, Hung-Yu Tseng, Hsin-Ying Lee, Maneesh Kumar Singh, Ming-Hsuan Yang
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 |
|---|---|---|---|
| Adding a pretrained ImageNet training (vs. untrained initialization) to the network | VGG-19, ResNet50, EfficientNet with zeros/circular/reflect/replicate/randn padding · Satellite image PPP measurement | SNR-PPP | Improved |
| Using randn padding instead of other padding schemes | VGG16-SOD on DUTS · DUTS saliency detection | weighted F-measure: 0.2570 vs 0.6269 (zeros) | Hurt |
| Using randn padding instead of zeros padding (ImageNet pretrained) | VGG-19, ResNet50, EfficientNet · ImageNet classification / PPP SNR | SNR-PPP: VGG-19 1.2366 vs 1.2289; ResNet50 0.6648 vs 0.8171; EfficientNet 0.5686 vs 0.7590 | No clear effect |
| Using no-padding instead of padding (ImageNet pretrained VGG-19) (ablation) | VGG-19 · ImageNet top-1 | Top-1 accuracy: 62.0396 vs 74.0972 (zeros) | Hurt |
| Replacing PosENet SPC with PPP SNR as the positional-information metric | Metric comparison across VGG-19, VGG16-SOD, ResNet50, EfficientNet · Standard deviation across five pretrained models | std of metric | Improved |
| Using no-padding model in metric measurement (PosENet SPC shows positive response; PPP is zero by definition) | PosENet vs PPP metrics · No-padding models | SPC / SNR-PPP response | Improved |
| Adding a principal point shift correction for stride-2 convolution on even-sized features | ResNet50 PPP measurement · PPP visualization | checkerboard artifacts in PPP map | Improved |
Closest papers by meaning
- Parabolic Position Encoding: Vision-Centric, Principled, Extrapolatable, General (NeurIPS 2026 · similarity 0.83)
- Cameras as Relative Positional Encoding (NeurIPS 2025 · similarity 0.83)
- Revisiting Multimodal Positional Encoding in Vision–Language Models (ICLR 2026 · similarity 0.82)
- DAPE: Data-Adaptive Positional Encoding for Length Extrapolation (NeurIPS 2024 · similarity 0.82)
- On the Emergence of Position Bias in Transformers (ICML 2025 · similarity 0.82)
- Circle-RoPE: Cone-like Decoupled Rotary Positional Embedding for Vision-Language Models (ICML 2026 · 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.
