Research map / Architectures and efficiency / CNN design
Transparent Vision: A Theory of Hierarchical Invariant Representations
Shuren Qi, Yushu Zhang, Chao Wang, Zhihua Xia, Xiaochun Cao, Fenglei Fan
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 scattering network with HIR | invariant representation · Digit benchmark (translation and rotation) | F1 (%): 56.23 -> 94.98 | Improved |
| add data augmentation | AlexNet · Digit benchmark (translation and rotation) | F1 (%): 64.27 -> 92.93 | Improved |
| add data augmentation | VGGNet · Digit benchmark (original) | F1 (%): 100 -> 95.68 | Hurt |
| add data augmentation | SimpleNet · Digit benchmark (original) | F1 (%): 98.60 -> 48.82 | Hurt |
| use over-complete wavelet instead of complete cosine representation | classical representations · Digit benchmark (original) | F1 (%): 45.43 -> 66.75 | Improved |
| swap SVM for NN classifier | Cosine features · AIGC forensics, 8 generators (average) | F1 (%): 98.03 -> 8.14 | Hurt |
| test with random orientation and flipping | VGGNet · AIGC forensics, 8 generators (average) | F1 (%): 93.66 -> 89.07 | Hurt |
| test with random orientation and flipping | HIR SVM · AIGC forensics, 8 generators (average) | F1 (%): 98.68 -> 98.63 | No clear effect |
| test with random orientation and flipping | ScatterNet SVM · AIGC forensics, 8 generators (average) | F1 (%): 95.77 -> 91.66 | Hurt |
| reduce training fraction from 5/5 to 1/9 | DenseNet · AIGC hybrid benchmark | F1 (%): 86.02 -> 76.10 | Hurt |
| reduce training fraction from 5/5 to 1/9 | HIR SVM · AIGC hybrid benchmark | F1 (%): 96.64 -> 94.73 | Hurt |
| use HIR instead of AIGC-specific forensic method ECCV'20 | AIGC forensics · AIGC hybrid benchmark (1/9) | F1 (%): 79.67 -> 94.73 | Improved |
Closest papers by meaning
- Understanding Image Representations by Measuring Their Equivariance and Equivalence (IJCV 2018 · similarity 0.87)
- Stretching Beyond the Obvious: A Gradient-Free Framework to Unveil the Hidden Landscape of Visual Invariance (ICLR 2026 · similarity 0.87)
- Exploiting Invariance in Training Deep Neural Networks (AAAI 2022 · similarity 0.87)
- On Making $SE(2)$-Invariant Networks Optimal (NeurIPS 2026 · similarity 0.86)
- PIEs: Pose Invariant Embeddings (CVPR 2019 · similarity 0.86)
- Discriminating image representations with principal distortions (ICLR 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.
