Research map / Detection and segmentation / CNN features
Reverse Perspective Network for Perspective-Aware Object Counting
Yifan Yang, Guorong Li, Zhe Wu, Li Su, Qingming Huang, Nicu Sebe
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 reverse perspective correction and adversarial residual learning | CSRNet · ShanghaiTech Part A | MAE: 68.2 -> 61.2 | Improved |
| add reverse perspective correction and adversarial residual learning | CSRNet · ShanghaiTech Part B | MAE: 10.6 -> 8.1 | Improved |
| add reverse perspective correction and adversarial residual learning | CSRNet · WorldExpo10 | MAE: 8.6 -> 8.2 | Improved |
| add reverse perspective correction and adversarial residual learning | CSRNet · UCSD | MSE: 1.47 -> 1.23 | Improved |
| add reverse perspective correction and adversarial residual learning | CSRNet · UAVVC | MAE: 18.32 -> 13.21 | Improved |
| add the residual adversarial network (ablation) | reverse perspective network with CSRNet on ShanghaiTech Part A · ShanghaiTech Part A | MAE: 63.4 -> 61.2 | Improved |
| train the reverse perspective network end-to-end with the regressor instead of pre-training it (ablation) | CSRNet on ShanghaiTech Part A · ShanghaiTech Part A | MAE: 61.2 -> 72.1 | Hurt |
| warp features instead of input images (ablation) | reverse perspective network with CSRNet on ShanghaiTech Part A · ShanghaiTech Part A | MAE: 61.2 -> 66.2 | Hurt |
| train the reverse perspective network with SGD instead of meta-learning (ablation) | CSRNet on ShanghaiTech Part A · ShanghaiTech Part A | MAE: 61.2 -> 70.6 | Hurt |
| use the ResNet-101 backbone with the proposed perspective method (ablation) | ResNet-101 regression backbone on ShanghaiTech Part A · ShanghaiTech Part A | MAE: 109.6 -> 94.3 | Improved |
| add reverse perspective correction to a ResNet-101 backbone (ablation) | ResNet-101 regression backbone on ShanghaiTech Part A · ShanghaiTech Part A | MAE: 68.2 -> 94.3 | Hurt |
Closest papers by meaning
- Towards Perspective-Free Object Counting with Deep Learning (ECCV 2016 · similarity 0.86)
- Revisiting Perspective Information for Efficient Crowd Counting (CVPR 2019 · similarity 0.86)
- A Large Contextual Dataset for Classification, Detection and Counting of Cars with Deep Learning (ECCV 2016 · similarity 0.84)
- Perspective-Guided Convolution Networks for Crowd Counting (ICCV 2019 · similarity 0.84)
- Drone-Based Object Counting by Spatially Regularized Regional Proposal Network (ICCV 2017 · similarity 0.84)
- Enhancing Low-Rank Adaptation with Recoverability-Based Reinforcement Pruning for Object Counting (AAAI 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.
