Research map / Architectures and efficiency / CNN design
InterActive: Inter-Layer Activeness Propagation
Lingxi Xie, Liang Zheng, Jingdong Wang, Alan L. Yuille, Qi Tian
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 |
|---|---|---|---|
| propagate final-layer activeness to low- and mid-level features | VGGNet features · pool-1 through pool-4, six classification datasets | accuracy | Improved |
| use final-layer rather than next-layer activeness supervision (ablation) | InterActive · Caltech256, Indoor-67, SUN-397, Pet-37, Flower-102, Bird-200 | accuracy: 84.64 -> 85.06; 78.23 -> 78.65; 62.50 -> 62.97; 93.22 -> 93.45; 96.26 -> 96.40; 74.61 -> 75.62 | Improved |
| use p=2 rather than p=1 in final-layer activeness (ablation) | InterActive · Caltech256, Indoor-67, SUN-397, Pet-37, Flower-102, Bird-200 | accuracy: 84.94 -> 85.06; 78.40 -> 78.65; 62.69 -> 62.97; 93.40 -> 93.45; 96.35 -> 96.40; 75.47 -> 75.62 | Improved |
| replace InterActive weighting with original average-pooled features (ablation) | InterActive concatenated features · Caltech256, Indoor-67, SUN-397, Pet-37, Flower-102, Bird-200 | accuracy: 84.02 -> 85.06; 78.02 -> 78.65; 62.30 -> 62.97; 93.02 -> 93.45; 95.70 -> 96.40; 73.35 -> 75.62 | Improved |
| replace InterActive weighting with original max-pooled features (ablation) | InterActive concatenated features · Caltech256, Indoor-67, SUN-397, Pet-37, Flower-102, Bird-200 | accuracy: 84.38 -> 85.06; 77.32 -> 78.65; 61.87 -> 62.97; 93.20 -> 93.45; 95.98 -> 96.40; 74.76 -> 75.62 | Improved |
| apply InterActive to high-level features | VGGNet pool-5 and fc-6 features · scene recognition | accuracy | Hurt |
| apply InterActive to high-level features | VGGNet pool-5 and fc-6 features · object recognition | accuracy | Mixed |
| use average pooling instead of max pooling (ablation) | original deep features · scene recognition | accuracy | Improved |
| use max pooling instead of average pooling (ablation) | original deep features · Bird-200 | accuracy | Improved |
| preserve image aspect ratio instead of warping input images (ablation) | VGGNet deep-feature extraction · Caltech256, SUN-397, Flower-102, fc-6 | accuracy: 80.41% -> 83.51%; 53.06% -> 61.30%; 84.89% -> 93.54% | Improved |
| apply InterActive response update to fc-7 and recompute fc-8 scores | VGG16, VGG19, and combined VGG models · ILSVRC2012 validation | error rate: 7.1% -> 6.8%; 7.0% -> 6.7%; 6.7% -> 6.5% | Improved |
| add object bounding boxes (ablation) | original features and InterActive features · Bird-200 | accuracy: 76.95% -> 77.53% | Improved |
Closest papers by meaning
- Discovering Influential Neuron Path in Vision Transformers (ICLR 2025 · similarity 0.84)
- Parallel Backpropagation for Shared-Feature Visualization (NeurIPS 2024 · similarity 0.84)
- From Activation to Initialization: Scaling Insights for Optimizing Neural Fields (CVPR 2024 · similarity 0.84)
- Neuron Activation Coverage: Rethinking Out-of-distribution Detection and Generalization (ICLR 2024 · similarity 0.84)
- NeurFlow: Interpreting Neural Networks through Neuron Groups and Functional Interactions (ICLR 2025 · similarity 0.83)
- LINE: LLM-based Iterative Neuron Explanations for Vision Models (NeurIPS 2026 · 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.
