Research map / 3D vision / Human pose estimation
Viewpoint Estimation---Insights & Model
Gilad Divon, Ayellet Tal
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 |
|---|---|---|---|
| replace VGG with ResNet backbone in a unified Faster R-CNN network (ablation) | VGG classification model · PASCAL3D+ (synthetic+real) | mAVP24: 36.1 -> 39.5 | Improved |
| replace VGG with ResNet backbone (real data only) (ablation) | VGG classification model · PASCAL3D+ (real data) | mAVP24: 31.1 -> 37.6 | Improved |
| use geometric loss instead of classification loss (ablation) | ResNet/Faster R-CNN · PASCAL3D+ | mAVP24: 39.5 -> 40.6 | Improved |
| integrate over viewpoints within a bin instead of selecting the max-activation bin (ablation) | ResNet/Faster R-CNN with geometric loss · PASCAL3D+ | mAVP24: 40.6 -> 43.2 | Improved |
| replace geometric loss with the viewpoint loss (triplet + flip) (ablation) | ResNet/Faster R-CNN with integral · PASCAL3D+ | mAVP24: 43.2 -> 44.4 | Improved |
| add unlabeled video data in semi-supervised triplet training (ablation) | viewpoint-loss model · PASCAL3D+ | mAVP24: 44.4 -> 45.9 | Improved |
| add synthetic images to real training data (ablation) | ResNet/Faster R-CNN classification · PASCAL3D+ | mAVP24: 37.6 -> 39.5 | Improved |
| use contrastive loss instead of geometric loss (ablation) | geometric-loss model · PASCAL3D+ (real data, loss comparison) | mAVP24: 43.2 -> 42.5 | Hurt |
| use flip loss instead of geometric loss (ablation) | geometric-loss model · PASCAL3D+ (real data, loss comparison) | mAVP24: 43.2 -> 43.6 | Improved |
| add triplet loss to geometric loss (ablation) | geometric-loss model · PASCAL3D+ (real data, loss comparison) | mAVP24: 43.2 -> 44.1 | Improved |
| add flip loss to triplet+geometric loss (viewpoint loss) (ablation) | triplet+geometric model · PASCAL3D+ (real data, loss comparison) | mAVP24: 44.1 -> 44.4 | Improved |
| full model versus prior state of the art | VGG/Fast R-CNN prior work · PASCAL3D+ | mAVP24: 36.1 -> 45.9 | Improved |
Closest papers by meaning
- Novel Object Viewpoint Estimation Through Reconstruction Alignment (CVPR 2020 · similarity 0.88)
- CNN^{2}: Viewpoint Generalization via a Binocular Vision (NeurIPS 2019 · similarity 0.86)
- Recognizing Objects From Any View With Object and Viewer-Centered Representations (CVPR 2020 · similarity 0.85)
- Self-Supervised Viewpoint Learning From Image Collections (CVPR 2020 · similarity 0.85)
- Cylindrical Convolutional Networks for Joint Object Detection and Viewpoint Estimation (CVPR 2020 · similarity 0.84)
- ViewNet: Unsupervised Viewpoint Estimation From Conditional Generation (ICCV 2021 · similarity 0.84)
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.
