Research map / Architectures and efficiency / CNN design
Top-Down Learning for Structured Labeling with Convolutional Pseudoprior
Saining Xie, Xun Huang, Zhuowen Tu
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 |
|---|---|---|---|
| add ConvPP | FCN-8s · Pascal-Context | mean IU: 37.8 -> 41.0 | Improved |
| add ConvPP | FCN-32s · Pascal-Context | mean IU: 35.1 -> 37.1 | Improved |
| add ConvPP | FCN-16s · Pascal-Context | mean IU: 37.6 -> 40.3 | Improved |
| add multi-scale context (pool5 + pool4 + pool3 instead of pool5 only) (ablation) | ConvPP · Pascal-Context | mean IU: 37.1 -> 41.0 | Improved |
| add ConvPP | FCN-8s · SIFT Flow | mean IU: 39.5 -> 40.7 | Improved |
| add ConvPP | FCN-16s · SIFT Flow | mean IU: 39.1 -> 39.7 | Improved |
| use 11x11 donut filter instead of 7x7 (ablation) | ConvPP · Pascal-Context | mean IU: 40.3 -> 41.0 | Improved |
| use 11x11 donut filter instead of 7x7 (ablation) | ConvPP · SIFT Flow | mean IU: 40.7 -> 32.4 | Hurt |
| jointly fine-tune the bottom-up FCN-8s with ConvPP (ablation) | ConvPP-8s · Pascal-Context | mean IU: 41.0 -> 40.89 | No clear effect |
| add ConvPP to the network (ablation) | NN without ConvPP · OCR (small training set) | error rate (%): 15.73 -> 6.49 | Improved |
| add ConvPP to the network (ablation) | NN without ConvPP · OCR (large training set) | error rate (%): 3.69 -> 0.57 | Improved |
| add ConvPP to the network (ablation) | NN without ConvPP · FAQ sentence labeling | error rate (%): 5.25 -> 1.09 | Improved |
Closest papers by meaning
- More Reliable Pseudo-labels, Better Performance: A Generalized Approach to Single Positive Multi-label Learning (ICCV 2025 · similarity 0.86)
- SemPPL: Predicting Pseudo-Labels for Better Contrastive Representations (ICLR 2023 · similarity 0.86)
- Detect, Replace, Refine: Deep Structured Prediction for Pixel Wise Labeling (CVPR 2017 · similarity 0.85)
- Label-Efficient Dataset Pruning via Semi-Supervised Pseudo-Labeling (NeurIPS 2026 · similarity 0.84)
- Learning from True-False Labels via Multi-modal Prompt Retrieving (ICML 2025 · similarity 0.84)
- Fusing Conditional Submodular GAN and Programmatic Weak Supervision (AAAI 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.
