Research map / Reinforcement learning / MDPs and planning
Weighing Counts: Sequential Crowd Counting by Reinforcement Learning
Liang Liu, Hao Lu, Hongwei Zou, Haipeng Xiong, Zhiguo Cao, Chunhua Shen
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 guiding reward (ablation) | LibraNet · ShanghaiTech Part A | MAE: 149.8 -> 55.9 | Improved |
| add force ending reward (ablation) | LibraNet · ShanghaiTech Part A | MAE: 62.7 -> 55.9 | Improved |
| add squeezing reward (ablation) | LibraNet · ShanghaiTech Part A | MAE: 63.5 -> 55.9 | Improved |
| Q-learning with designed rewards instead of imitation of optimal actions (ablation) | LibraNet · ShanghaiTech Part A | MAE: 64.7 -> 55.9 | Improved |
| increase maximum step from 4 to 8 (ablation) | LibraNet · ShanghaiTech Part A | MAE: 126.3 -> 55.9 | Improved |
| increase maximum step from 8 to 12 (ablation) | LibraNet · ShanghaiTech Part A | MAE: 55.9 -> 62.5 | Hurt |
| tolerance range 0.5 versus 0.1 (ablation) | LibraNet · ShanghaiTech Part A | MAE: 61.0 -> 55.9 | Improved |
| add LibraNet to ImageNet-pretrained classification baseline (ablation) | ImageNet Classification · ShanghaiTech Part A (plug-in) | MAE: 140.4 -> 119.7 | Improved |
| add LibraNet to ImageNet-pretrained regression baseline (ablation) | ImageNet Regression · ShanghaiTech Part A (plug-in) | MAE: 156.2 -> 126.6 | Improved |
| add LibraNet to TasselNetv2 (ablation) | TasselNetv2 · ShanghaiTech Part A (plug-in) | MAE: 68.6 -> 64.7 | Improved |
| add LibraNet to blockwise classification (ablation) | Blockwise Classification · ShanghaiTech Part A (plug-in) | MAE: 62.8 -> 55.9 | Improved |
| sequential decision (LibraNet) versus BCNet on patch-level counting | BCNet · ShanghaiTech Part A | GAME3: 116.7 -> 113.1 | Improved |
Closest papers by meaning
- To Choose or to Fuse? Scale Selection for Crowd Counting (AAAI 2021 · similarity 0.85)
- Context-Aware Crowd Counting (CVPR 2019 · similarity 0.84)
- Crowd Counting via Adversarial Cross-Scale Consistency Pursuit (CVPR 2018 · similarity 0.84)
- Free Lunch Enhancements for Multi-modal Crowd Counting (CVPR 2025 · similarity 0.84)
- Counting without Scale: Scale-Consistent Error Correction for Crowd Counting (NeurIPS 2026 · similarity 0.84)
- ART for Diffusion Sampling: A Reinforcement Learning Approach to Timestep Scheduling (NeurIPS 2026 · similarity 0.83)
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.
