Research map / Detection and segmentation / CNN features
Hierarchical Spherical CNNs With Lifting-Based Adaptive Wavelets for Pooling and Unpooling
Mingxing Xu, Chenglin Li, Wenrui Dai, Siheng Chen, Junni Zou, Pascal Frossard, Hongkai Xiong
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 downsampling pooling with learned lifting-based pooling and unpooling (ablation) | UGSCNN-style backbone · SCIFAR10 reconstruction (test) | MSE: 0.0030 -> 0.0017 | Improved |
| add lifting-based pooling with padding unpooling (ablation) | HS-CNN · SCIFAR10 reconstruction (test) | MSE: 0.0030 -> 0.0025 | Improved |
| add lifting-based unpooling (inverse wavelet) on top of lifting pooling (ablation) | LiftHS-CNN* · SCIFAR10 reconstruction (test) | MSE: 0.0025 -> 0.0017 | Improved |
| use learned adaptive wavelets instead of handcrafted wavelets (ablation) | LiftHS-CNN · SCIFAR10 reconstruction (test) | MSE: 0.0021 -> 0.0017 | Improved |
| use lifting pooling instead of mean pooling (ablation) | UGSCNN-style backbone · SMNIST | accuracy (%): 99.55 -> 99.60 | Improved |
| use lifting pooling instead of downsampling (ablation) | UGSCNN-style backbone · SModelNet40 | accuracy (%): 89.26 -> 91.37 | Improved |
| use learned instead of handcrafted wavelets (ablation) | LiftHS-CNN · SModelNet40 | accuracy (%): 90.96 -> 91.37 | Improved |
| use mean or max pooling instead of downsampling (ablation) | UGSCNN-style backbone · SModelNet40 | accuracy (%): 89.26 -> 90.76 | Improved |
| use lifting pooling and unpooling instead of downsampling (ablation) | UGSCNN-style backbone · Stanford 2D3DS segmentation | mIoU: 0.3789 -> 0.4534 | Improved |
| use learned instead of handcrafted wavelets (ablation) | LiftHS-CNN · Stanford 2D3DS segmentation | mIoU: 0.4217 -> 0.4534 | Improved |
| use lifting-based pooling/unpooling instead of downsampling/padding | UGSCNN · Stanford 2D3DS segmentation | mAcc (%): 54.65 -> 61.87 | Improved |
| use handcrafted spherical wavelets instead of mean pooling (ablation) | UGSCNN-style backbone · SMNIST | accuracy (%): 99.55 -> 99.50 | No clear effect |
Closest papers by meaning
- Scattering Networks on the Sphere for Scalable and Rotationally Equivariant Spherical CNNs (ICLR 2022 · similarity 0.88)
- Spin-Weighted Spherical CNNs (NeurIPS 2020 · similarity 0.87)
- Scaling Spherical CNNs (ICML 2023 · similarity 0.87)
- Efficient Generalized Spherical CNNs (ICLR 2021 · similarity 0.87)
- Spherical CNNs (ICLR 2018 · similarity 0.87)
- Attention on the Sphere (NeurIPS 2025 · similarity 0.83)
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.
