Research map / Detection and segmentation / CNN features
End-To-End People Detection in Crowded Scenes
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use recurrent joint decoding instead of OverFeat regression with non-maximum suppression | People detection with GoogLeNet features · Crowded-scenes test set | AP: $0.67$ -> $0.78$ | Helped |
| Use recurrent joint decoding instead of OverFeat regression with non-maximum suppression | People detection with GoogLeNet features · Crowded-scenes test set | recall: $71\%$ -> $81\%$ | Helped |
| Use recurrent joint decoding instead of OverFeat regression with non-maximum suppression | People detection with GoogLeNet features · Crowded-scenes test set | COUNT: $1.05$ -> $0.76$ | Helped |
| Use fixed-order matching instead of Hungarian matching (ablation) | End-to-End People Detection · Crowded scenes | AP: 0.78 -> 0.60 AP | Hurt |
| Use Hungarian matching with overlap terms instead of matching only the first k predictions (ablation) | End-to-End People Detection · Crowded scenes | confidence calibration | Helped |
| Tie output-to-candidate weights across LSTM decoding steps (ablation) | End-to-End People Detection · Crowded-scenes validation set | AP: $.85$ -> $.82$ | Hurt |
| Fine-tune GoogLeNet features for the decoder (ablation) | End-to-End People Detection · Crowded scenes | AP: reduced AP by $.29$ without fine tuning | Helped |
| Add dropout with probability 0.15 to each LSTM output (ablation) | End-to-End People Detection · Crowded scenes | AP: reduces AP by $.011$ when removed | Helped |
| Remove L2 regularization (ablation) | End-to-End People Detection · Crowded scenes | trainability | Helped |
| Apply L2 regularization with multiplier 1e-6 to GoogLeNet (ablation) | End-to-End People Detection · Crowded-scenes validation set | AP: reduced AP by $.03$ | Hurt |
| Increase the overcomplete prediction set beyond 5 predictions (ablation) | End-to-End People Detection · Crowded scenes | AP | No clear effect |
| Feed image features at every LSTM step rather than only to the first step (ablation) | End-to-End People Detection · Crowded scenes | performance | No clear effect |
Closest papers by meaning
- Progressive End-to-End Object Detection in Crowded Scenes (CVPR 2022 · similarity 0.88)
- Detection in Crowded Scenes: One Proposal, Multiple Predictions (CVPR 2020 · similarity 0.87)
- Context-Aware Crowd Counting (CVPR 2019 · similarity 0.87)
- Divide and Grow: Capturing Huge Diversity in Crowd Images With Incrementally Growing CNN (CVPR 2018 · similarity 0.86)
- Adaptive NMS: Refining Pedestrian Detection in a Crowd (CVPR 2019 · similarity 0.85)
- CrowdCLIP: Unsupervised Crowd Counting via Vision-Language Model (CVPR 2023 · 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.