Research map / Point clouds / Segmentation and detection
Shifted Autoencoders for Point Annotation Restoration in Object Counting
Yuda Zou, Xin Xiao, Peilin Zhou, Zhichao Sun, Bo Du, Yongchao Xu
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 |
|---|---|---|---|
| train with SAE-restored point annotations | P2PNet · SH PartA | MAE: 52.8 -> 48.2 | Improved |
| train with SAE-restored point annotations | BL · JHU-Crowd++ | MAE: 74.4 -> 60.8 | Improved |
| train with SAE-restored point annotations | ASPDNet · RSOC_Small-vehicle | MAE: 252.6 -> 182.0 | Improved |
| train with SAE-restored point annotations | SAUNet · DCC | MAE / MSE: 3.0 -> 2.6 MAE, 4.8 -> 3.4 MSE | Improved |
| train with SAE-restored point annotations | anti-noise methods (BL, NoiseCC, RSI) · crowd counting datasets | MAE / MSE: average -4.1 MAE and -11.5 MSE | Improved |
| increase sampling radius factor alpha from 0.4 to 0.6 (ablation) | BL & SAE · UCF-QNRF | MAE: 81.6 -> 91.0 | Hurt |
| reduce alpha from 0.4 to 0.3 (ablation) | BL & SAE · UCF-QNRF | MAE: 81.6 -> 86.5 | Hurt |
| use SAE pretrained backbone in addition to restored annotations (ablation) | P2PNet · SH PartA | MAE: 48.2 -> 47.2 | Improved |
| use only SAE pretrained backbone, without restored annotations (ablation) | P2PNet · SH PartA | MAE: 52.8 -> 49.7 | Improved |
| apply SAE restoration to noisy annotations | BL · UCF-QNRF with synthetic offset noise | MAE | Improved |
| train with SAE-restored point annotations | Toy stick-figure heads · simulated annotations | position variance: 0.223 x r^2 -> 0.016 x r^2 | Improved |
Closest papers by meaning
- Learning to Count without Annotations (CVPR 2024 · similarity 0.87)
- A Fixed-Point Approach to Unified Prompt-Based Counting (AAAI 2024 · similarity 0.87)
- OmniCount: Multi-label Object Counting with Semantic-Geometric Priors (AAAI 2025 · similarity 0.85)
- CountGD: Multi-Modal Open-World Counting (NeurIPS 2024 · similarity 0.85)
- COUNTS: Benchmarking Object Detectors and Multimodal Large Language Models under Distribution Shifts (CVPR 2025 · similarity 0.85)
- T2ICount: Enhancing Cross-modal Understanding for Zero-Shot Counting (CVPR 2025 · similarity 0.85)
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.
