Research map / 3D vision / Face and shape
Robust Attentional Aggregation of Deep Feature Sets for Multi-view 3D Reconstruction
Bo Yang, Sen Wang, Andrew Markham, Niki Trigoni
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 |
|---|---|---|---|
| train with FASet two-stage algorithm instead of joint end-to-end training (ablation) | AttSets · ShapeNet-r2n2 Group 4 (1 view) | mean IoU: 0.307 -> 0.642 | Improved |
| train with FASet instead of joint training (ablation) | AttSets · ShapeNet-r2n2 Group 4 (24 views) | mean IoU: 0.680 -> 0.695 | Improved |
| use feature-wise instead of element-wise attention scores (ablation) | AttSets · ShapeNet-r2n2 Group 4 (1 and 24 views) | mean IoU: 24 views: 0.672 -> 0.695 | Improved |
| use element-wise attention instead of feature-wise (ablation) | AttSets · ShapeNet-r2n2 Group 4 (2 views) | mean IoU: 0.660 -> 0.653 | Hurt |
| use fc attention instead of conv2d attention (ablation) | AttSets · ShapeNet-r2n2 Group 4 (24 views) | mean IoU: 0.676 -> 0.695 | Improved |
| use fc instead of conv3d attention (ablation) | AttSets · ShapeNet-r2n2 Group 4 (24 views) | mean IoU: 0.690 -> 0.695 | No clear effect |
| replace GRU with AttSets | Base_r2n2 · ShapeNet-r2n2 Group 5 (24 views) | mean IoU: 0.651 -> 0.694 | Improved |
| replace mean pooling with AttSets | Base_r2n2 · ShapeNet-r2n2 Group 5 (1 view) | mean IoU: 0.598 -> 0.642 | Improved |
| replace max pooling with AttSets | Base_r2n2 · ShapeNet-lsm (1 view) | mean IoU: 0.276 -> 0.404 | Improved |
| replace max pooling with AttSets | Base_r2n2 · ModelNet40 Group 2 (1 view) | mean IoU: 0.461 -> 0.487 | Improved |
| replace max pooling with AttSets | Base_silnet · Blobby Group 2 (4 views) | mean IoU: 0.929 -> 0.937 | Improved |
| fine-tune pooling baselines with random N in [1,24] instead of fixed N=24 (ablation) | max pooling · ShapeNet-r2n2 (1 view) | mean IoU: 0.504 -> 0.601 | Improved |
Closest papers by meaning
- Multiview Aggregation for Learning Category-Specific Shape Reconstruction (NeurIPS 2019 · similarity 0.88)
- Multi-View 3D Reconstruction With Transformers (ICCV 2021 · similarity 0.87)
- MVTN: Multi-View Transformation Network for 3D Shape Recognition (ICCV 2021 · similarity 0.87)
- MVTN: Learning Multi-view Transformations for 3D Understanding (IJCV 2024 · similarity 0.87)
- Image-Based 3D Object Reconstruction: State-of-the-Art and Trends in the Deep Learning Era (TPAMI 2019 · similarity 0.87)
- Implicit Functions in Feature Space for 3D Shape Reconstruction and Completion (CVPR 2020 · 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.
