Research map / Detection and segmentation / CNN features
Mixture Dense Regression for Object Detection and Human Pose Estimation
Ali Varamesh, Tinne Tuytelaars
Check 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 |
|---|---|---|---|
| Replace single-mode box regression with a 3-component mixture density network | CenterNet HG · COCO-val object detection | AP: 35.9 -> 39.0 | Improved |
| Replace single-mode box regression with a 3-component mixture density network | CenterNet HG · COCO-test-dev object detection | AP: 0.401 -> 0.416 | Improved |
| Replace single-mode box regression with a 3-component mixture density network | CenterNet DLA34 · COCO-test-dev object detection | AP: 0.386 -> 0.406 | Improved |
| Replace single-mode offset regression with a 2-component mixture density network | CenterNet HG · COCO-val human pose estimation | AP: 46.4 -> 52.3 | Improved |
| Replace single-mode offset regression with a 3-component mixture density network | CenterNet HG · COCO-test-dev human pose estimation | AP: 55.6 -> 57.9 | Improved |
| Use 5 rather than 3 mixture components (ablation) | MDN object detection · COCO-val object detection | AP: 39.0 -> 38.8 | No clear effect |
| Use more than two mixture components (ablation) | MDN human pose estimation · COCO human pose estimation | precision | No clear effect |
| Use more than two mixture components (ablation) | MDN human pose estimation · COCO human pose estimation | recall: < 1% | Improved |
| Give each object category a separate box prediction head (ablation) | Single-mode base object detector · COCO object detection | AP | No clear effect |
| Choose the maximum-scoring mixture component rather than mixing component outputs (ablation) | MDN dense regression at inference · COCO object detection and human pose estimation | AP | Improved |
| Raise the minimum allowed variance activation value above one (ablation) | MDN variance activation · COCO-val | AP | No clear effect |
| Use mixture regression instead of single-mode regression | Dense object detection · COCO-test-dev object detection | FPS: 8.44 -> 7.47 | Hurt |
Closest papers by meaning
- Generating Multiple Hypotheses for 3D Human Pose Estimation With Mixture Density Network (CVPR 2019 · similarity 0.88)
- Active Learning for Deep Object Detection via Probabilistic Modeling (ICCV 2021 · similarity 0.87)
- DensePose: Dense Human Pose Estimation in the Wild (CVPR 2018 · similarity 0.87)
- SimPose: Effectively Learning DensePose and Surface Normals of People from Simulated Data (ECCV 2020 · similarity 0.86)
- ObjectNet: A large-scale bias-controlled dataset for pushing the limits of object recognition models (NeurIPS 2019 · similarity 0.86)
- Training Multi-Object Detector by Estimating Bounding Box Distribution for Input Image (ICCV 2021 · similarity 0.86)
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.
