Research map / Detection and segmentation / Open-set and few-shot detection
Exploring Tuning Characteristics of Ventral Stream’s Neurons for Few-Shot Image Classification
Lintao Dong, Wei Zhai, Zheng-Jun Zha
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 |
|---|---|---|---|
| Apply HFR and foreground segmentation | RENet and Distribution Calibration variants · CUB, miniImageNet, CIFAR-FS, and tieredImageNet; 5-way 1-shot and 5-shot | accuracy: RENet: +3.84, +1.86, +2.11, +1.65, +1.31, +0.85, +2.33, +1.87; LR-DC: +2.70, +1.93, +1.41, +1.34, +1.07, +0.74, +1.29, + | Improved |
| Add shape-cue neuron regularization (ablation) | RENet · CUB and miniImageNet 5-way 1-shot | accuracy: +1.27; +1.13 | Improved |
| Add color-cue neuron regularization after shape cues (ablation) | RENet · CUB and miniImageNet 5-way 1-shot | accuracy: 80.83 (+1.27) -> 81.17 (+1.61); 68.73 (+1.13) -> 68.98 (+1.38) | Improved |
| Add foreground segmentation for belongingness after shape and color cues (ablation) | RENet · CUB and miniImageNet 5-way 1-shot | accuracy: 81.17 (+1.61) -> 82.26 (+2.70); 68.98 (+1.38) -> 69.71 (+2.11) | Improved |
| Add V2 and V4 regularization after V1 regularization (ablation) | RENet · CUB and miniImageNet 5-way 1-shot | accuracy: 81.76 (+2.27) -> 82.57 (+3.08) -> 83.14 (+3.65); 68.21 (+0.61) -> 68.67 (+1.07) -> 68.98 (+1.38) | Improved |
| Reverse the V1→V2→V4 regularization order (ablation) | RENet HFR · CUB and miniImageNet 5-way 1-shot | accuracy: 83.14 -> 82.17; 68.98 -> 68.34 | Hurt |
| Concatenate neuron responses with model input (ablation) | RENet · CUB and miniImageNet 5-way 1-shot | accuracy: 79.49 -> 77.65; 67.60 -> 66.13 | Hurt |
| Concatenate neuron responses with features before the fully connected layer (ablation) | RENet · CUB and miniImageNet 5-way 1-shot | accuracy: 79.49 -> 79.11; 67.60 -> 67.26 | Hurt |
| Transfer neuron-model knowledge by distillation (ablation) | RENet · CUB and miniImageNet 5-way 1-shot | accuracy: 79.49 -> 81.01; 67.60 -> 68.14 | Improved |
| Use hierarchical feature regularization instead of distillation (ablation) | RENet · CUB and miniImageNet 5-way 1-shot | accuracy: 81.01 -> 83.14; 68.14 -> 68.98 | Improved |
Closest papers by meaning
- Vision CNNs trained to estimate spatial latents learned similar ventral-stream-aligned representations (ICLR 2025 · similarity 0.87)
- Parallel Backpropagation for Shared-Feature Visualization (NeurIPS 2024 · similarity 0.85)
- Pushing the Limit of Fine-Tuning for Few-Shot Learning: Where Feature Reusing Meets Cross-Scale Attention (AAAI 2024 · similarity 0.85)
- STSBench: A Large-Scale Dataset for Modeling Neuronal Activity in the Dorsal Stream of Primate Visual Cortex (NeurIPS 2025 · similarity 0.85)
- Sparse components distinguish visual pathways & their alignment to neural networks (ICLR 2025 · similarity 0.84)
- L-WISE: Boosting Human Visual Category Learning Through Model-Based Image Selection and Enhancement (ICLR 2025 · similarity 0.84)
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.
