Research map / Detection and segmentation / CNN features
PT-CapsNet: A Novel Prediction-Tuning Capsule Network Suitable for Deeper Architectures
Chenbin Pan, Senem Velipasalar
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 |
|---|---|---|---|
| add capsule-wise prediction then feature-wise tuning instead of the reverse order (ablation) | PT-CapsNet small model · MNIST | testing error: 0.75 -> 0.63 | Improved |
| add capsule-wise prediction then feature-wise tuning instead of the reverse order (ablation) | PT-CapsNet small model · Fashion-MNIST | testing error: 8.29 -> 7.77 | Improved |
| add generated logits (extra FC-PT-Caps layer) instead of L2-norm logits (ablation) | PT-CapsNet small model · MNIST | testing error: 88.65 -> 0.63 | Improved |
| use ReLU instead of Squash with generated logits (ablation) | PT-CapsNet small model · MNIST | testing error: 0.74 -> 0.63 | Improved |
| replace conv and FC layers with LC-PT-Caps and FC-PT-Caps | ResNet-110 · CIFAR-10 | top-1 accuracy: 92.69 -> 93.59 | Improved |
| replace conv and FC layers with LC-PT-Caps and FC-PT-Caps | WRN-28-10 · CIFAR-10 | top-1 accuracy: 95.95 -> 95.98 | No clear effect |
| replace conv and FC layers with LC-PT-Caps and FC-PT-Caps | DenseNet-100 · CIFAR-10 | top-1 accuracy: 95.33 -> 95.71 | Improved |
| replace conv and FC layers with PT-Caps layers | ResNet-110 · CIFAR-100 | top-1 accuracy: 70.32 -> 71.39 | Improved |
| replace conv and FC layers with PT-Caps layers | DenseNet-100 · CIFAR-100 | top-1 accuracy: 77.12 -> 78.36 | Improved |
| replace conv and FC layers with PT-Caps layers | ResNet-110 · FashionMNIST | top-1 accuracy: 94.36 -> 95.28 | Improved |
| replace conv layers with PT-Caps layers | U-Net · ISIC2018 | mIoU: 65.57 -> 66.86 | Improved |
| replace conv layers with PT-Caps layers | DeepLabv3+ · ISIC2018 | mIoU: 82.85 -> 83.12 | Improved |
Closest papers by meaning
- Why Capsule Neural Networks Do Not Scale: Challenging the Dynamic Parse-Tree Assumption (AAAI 2023 · similarity 0.90)
- DeepCaps: Going Deeper With Capsule Networks (CVPR 2019 · similarity 0.89)
- STAR-Caps: Capsule Networks with Straight-Through Attentive Routing (NeurIPS 2019 · similarity 0.89)
- Improving the Robustness of Capsule Networks to Image Affine Transformations (CVPR 2020 · similarity 0.89)
- CapProNet: Deep Feature Learning via Orthogonal Projections onto Capsule Subspaces (NeurIPS 2018 · similarity 0.88)
- OrthCaps: An Orthogonal CapsNet with Sparse Attention Routing and Pruning (CVPR 2024 · similarity 0.88)
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.
