Research map / Architectures and efficiency / CNN design
ReDro: Efficiently Learning Large-sized SPD Visual Representation
Saimunur Rahman, Lei Wang, Changming Sun, Luping Zhou
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 ReDro (k=4, EL-1) to DeepCOV | DeepCOV 512x512 · Birds | accuracy: 85.4 -> 86.5 | Improved |
| add ReDro (SL-6) to DeepCOV | DeepCOV 512x512 · Birds | accuracy: 85.4 -> 86.7 | Improved |
| add ReDro (k=4, EL-5) to DeepCOV-ResNet | DeepCOV-ResNet 1024x1024 · Cars | accuracy: 85.0 -> 89.8 | Improved |
| add ReDro k=4 to MPN-COV | MPN-COV · Airplane | accuracy: 86.1 -> 88.2 | Improved |
| add ReDro k=4 to iSQRT-COV | iSQRT-COV · Cars | accuracy: 92.6 -> 92.6 | No clear effect |
| add ReDro k=4 to IBCNN | IBCNN · Birds | accuracy: 85.4 -> 84.6 | Hurt |
| use ReDro k=8 instead of full covariance | DeepCOV 1024x1024 · matrix normalisation time | time (s): 0.097 -> 0.056 | Improved |
| use ReDro on small 256x256 matrix | DeepCOV 256x256 · matrix normalisation time | time (s): 0.013 -> 0.011 | Hurt |
| add ReDro to DeepCOV training | DeepCOV · Birds training time | hours: 52.4 -> 24.3 | Improved |
| use batch-level permutation every batch (BL-1) instead of no ReDro (ablation) | DeepCOV 512x512 · MIT Indoor | accuracy: 79.2 -> 76.8 | Hurt |
| increase group number from 4 to 16 (ablation) | DeepCOV-ResNet 1024x1024 · Birds | accuracy: 86.2 -> 82.9 | Hurt |
| increase group number from 2 to 4 (ablation) | DeepCOV-ResNet 512x512 · Birds | accuracy: 85.8 -> 85.9 | Improved |
Closest papers by meaning
- DeepKSPD: Learning Kernel-matrix-based SPD Representation for Fine-grained Image Recognition (ECCV 2018 · similarity 0.85)
- DropCov: A Simple yet Effective Method for Improving Deep Architectures (NeurIPS 2022 · similarity 0.85)
- Schur's Positive-Definite Network: Deep Learning in the SPD cone with structure (ICLR 2025 · similarity 0.84)
- Deep Rank-One Tensor Functional Factorization for Multi-Dimensional Data Recovery (AAAI 2025 · similarity 0.83)
- Navigating Beyond Dropout: An Intriguing Solution towards Generalizable Image Super Resolution (CVPR 2024 · similarity 0.83)
- Learning Structured Representations with Hyperbolic Embeddings (NeurIPS 2024 · 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.
