Research map / 3D vision / Human pose estimation
Pose Augmentation: Class-agnostic Object Pose Transformation for Object Recognition
Yunhao Ge, Jiaping Zhao, Laurent Itti
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 |
|---|---|---|---|
| use OPT-Net eliminate-add generator instead of StarGAN | pose transformation · iLab-20M pose transformation (6 yaw poses, foreground) | MSE: 502.51 -> 374.76 | Improved |
| use OPT-Net instead of StarGAN | pose transformation · iLab-20M pose transformation (6 yaw poses, foreground) | PSNR: 21.95 -> 23.04 | Improved |
| add OPT-Net synthesized images for missing poses | ResNet-18 on P-UB · iLab-20M vehicle recognition | overall accuracy: 52.26% -> 59.15% | Improved |
| add interpolated extra poses to synthesized balanced set (ablation) | ResNet-18 on S-P-B · iLab-20M vehicle recognition | overall accuracy: 59.2 -> 61.2 | Improved |
| use OPT-Net synthesis instead of traditional augmentation | ResNet-18 on P-UB · iLab-20M vehicle recognition | overall accuracy: 52.3 -> 59.2 | Improved |
| add OPT-Net synthesized poses on pretrained-and-finetuned model | AlexNet on RGB-D P-UB · RGB-D object recognition | overall accuracy: 99.1 -> 99.7 | Improved |
| add synthesized poses when only 1 pose per class is available | AlexNet on confusable RGB-D classes · RGB-D confusable categories | overall accuracy: 75.1 -> 83.2 | Improved |
| add synthesized poses when 2 poses per class are available | AlexNet on confusable RGB-D classes · RGB-D confusable categories | overall accuracy: 90.4 -> 94.2 | Improved |
| add synthesized poses when 3 poses per class are available | AlexNet on confusable RGB-D classes · RGB-D confusable categories | overall accuracy: 99.3 -> 99.4 | No clear effect |
| use OPT-Net instead of traditional augmentation with 1 pose per class | AlexNet on confusable RGB-D classes · RGB-D confusable categories | overall accuracy: 77.6 -> 83.2 | Improved |
| pretrain on OPT-Net synthesized images instead of StarGAN | ResNet-18 vehicle classifier · ImageNet car class | accuracy: 41.84 -> 76.20 | Improved |
| pretrain AlexNet on OPT-Net synthesized poses instead of scratch | AlexNet · ILSVRC-2010, 10 images per class | top-5 accuracy: 4.32 -> 10.94 | Improved |
Closest papers by meaning
- General Object Pose Transformation Network from Unpaired Data (ECCV 2022 · similarity 0.89)
- NOPE: Novel Object Pose Estimation from a Single Image (CVPR 2024 · similarity 0.88)
- RotationNet: Joint Object Categorization and Pose Estimation Using Multiviews From Unsupervised Viewpoints (CVPR 2018 · similarity 0.88)
- Deep Learning-Based Object Pose Estimation: A Comprehensive Survey (IJCV 2026 · similarity 0.88)
- DualPoseNet: Category-Level 6D Object Pose and Size Estimation Using Dual Pose Network With Refined Learning of Pose Consistency (ICCV 2021 · similarity 0.88)
- GIVEPose: Gradual Intra-class Variation Elimination for RGB-based Category-Level Object Pose Estimation (CVPR 2025 · 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.
