Research map / Architectures and efficiency / CNN design
Learning Bag-Of-Features Pooling for Deep Convolutional Neural Networks
Nikolaos Passalis, Anastasios Tefas
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 |
|---|---|---|---|
| train with images of several sizes (Train C) instead of only 28x28 (ablation) | CBoF (0, 32) · MNIST test | error (%) at 28x28: 0.88 -> 0.68 | Improved |
| train with images of several sizes (Train C) instead of only 28x28 (ablation) | CBoF (0, 32) · MNIST test 20x20 | error (%): 19.69 -> 1.44 | Improved |
| train with images of several sizes (Train C) instead of only 28x28 (ablation) | CBoF (0, 32) · MNIST test 36x36 | error (%): 25.39 -> 1.60 | Improved |
| train with images of several sizes (Train C) instead of only 227x227 (ablation) | CBoF (0, 256) · 15-scene | accuracy (%): 90.66 -> 92.38 | Improved |
| add spatial segmentation level 1 with 128 codewords vs level 0 (ablation) | CBoF · MNIST test | error (%): 0.74 -> 0.61 | Improved |
| add more RBF neurons (up to 128) without spatial segmentation (ablation) | CBoF · MNIST test | error (%): 1.13 -> 0.74 | Improved |
| add spatial segmentation (ablation) | CBoF · 15-scene | accuracy | No clear effect |
| add spatial segmentation (CBoF (1, 32) vs CBoF (0, 128), same parameters) (ablation) | CBoF · AFLW yaw (64x64) | angular error (deg): 16.23 -> 11.92 | Improved |
| use CBoF(1, 32) instead of SPP | SPP · MNIST test 20x20 | error (%): 1.78 -> 0.94 | Improved |
| use CBoF (0, 256) instead of CNN | CNN · 15-scene (227) | accuracy (%): 88.79 -> 91.38 | Improved |
| use CBoF(1, 32) instead of CNN | CNN · MNIST test (28) | error (%): 0.56 -> 0.55 | No clear effect |
| add a 1x1 convolutional layer with 64 filters to GMP (ablation) | GMP · MNIST test 20x20 | error (%): 3.22 -> 3.52 | Hurt |
Closest papers by meaning
- TI-Pooling: Transformation-Invariant Pooling for Feature Learning in Convolutional Neural Networks (CVPR 2016 · similarity 0.86)
- Cross-Convolutional-Layer Pooling for Image Recognition (TPAMI 2016 · similarity 0.86)
- Gaussian-Based Pooling for Convolutional Neural Networks (NeurIPS 2019 · similarity 0.86)
- DSConv: Efficient Convolution Operator (ICCV 2019 · similarity 0.85)
- Wavelet Pooling for Convolutional Neural Networks (ICLR 2018 · similarity 0.85)
- PICNN: A Pathway towards Interpretable Convolutional Neural Networks (AAAI 2024 · 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.
