Research map / Architectures and efficiency / CNN design
Principal Feature Visualisation in Convolutional Neural Networks
Marianne Bakken, Johannes Kvam, Alexey A. Stepanov, Asbjørn Berge
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 |
|---|---|---|---|
| weight upsampled principal feature map with summed earlier-layer activation maps (ablation) | PFV on VGG16 · dog and cat batch (qualitative) | visual background suppression | Improved |
| use a single-image batch instead of a mixed-class batch (ablation) | PFV · dog and cat images (qualitative) | feature granularity | Mixed |
| randomise network weights (untrained VGG16) (ablation) | PFV · sanity check, qualitative | visualisation similarity to trained model | Hurt |
| restrict the batch to five similar animal classes instead of 20 diverse classes (ablation) | PFV on VGG16 · Pascal VOC2012 (qualitative) | separation of cats and dogs from other animals | Improved |
| use weak PFV feature response as a predictor of poor fine-tuned accuracy | VGG16 fine-tuned on VOC2012 · chair, table, sofa classes | per-class accuracy: table 54, sofa 73, chair 73 versus cat 95 | Improved |
Closest papers by meaning
- Visualizing Deep Convolutional Neural Networks Using Natural Pre-images (IJCV 2016 · similarity 0.85)
- Design of Kernels in Convolutional Neural Networks for Image Classification (ECCV 2016 · similarity 0.84)
- PSConv: Squeezing Feature Pyramid into One Compact Poly-Scale Convolutional Layer (ECCV 2020 · similarity 0.84)
- Curriculum By Smoothing (NeurIPS 2020 · similarity 0.84)
- Contextual Convolutional Networks (ICLR 2023 · similarity 0.84)
- PICNN: A Pathway towards Interpretable Convolutional Neural Networks (AAAI 2024 · similarity 0.82)
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.
