Research map / Architectures and efficiency / CNN design
Locally-Transferred Fisher Vectors for Texture Classification
Yang Song, Fan Zhang, Qing Li, Heng Huang, Lauren J. O'Donnell, Weidong Cai
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 |
|---|---|---|---|
| transform FV-CNN with LFV network | FV-CNN · KTH-TIPS2 | accuracy (%): 81.4 -> 82.6 | Improved |
| transform FV-CNN with LFV network | FV-CNN · FMD | accuracy (%): 79.7 -> 82.1 | Improved |
| transform FV-CNN with LFV network | FV-CNN · DTD | accuracy (%): 72.4 -> 73.8 | Improved |
| fine-tune VGG-VD on texture dataset before FV encoding | FV-CNN · DTD | accuracy (%): 72.4 -> 71.8 | Hurt |
| learn FV descriptor end-to-end with an FV layer | FV-CNN · DTD | accuracy (%): 72.4 -> 68.2 | Hurt |
| concatenate transformed FC-CNN descriptor with LFV | LFV · FMD | accuracy (%): 82.1 -> 83.5 | Improved |
| concatenate transformed FC-CNN descriptor with LFV | LFV · DTD | accuracy (%): 73.8 -> 75.2 | Improved |
| concatenate transformed FC-CNN descriptor with LFV | LFV · KTH-TIPS2 | accuracy (%): 82.6 -> 83.1 | Improved |
| use LFV instead of compact bilinear pooling at 8192 dims | CBP-TS · DTD (1/3 train, 2/3 test) | accuracy (%): 67.8 -> 68.6 | Improved |
| use locally connected layer instead of fully connected layer (ablation) | LFV · DTD | accuracy (%) | Improved |
| use SVM hinge loss instead of softmax loss (ablation) | LFV · KTH-TIPS2 / FMD / DTD | accuracy (%): 0.5% lower accuracy on average with softmax | Improved |
| share weights across every 8 filters instead of 4 (P=8) (ablation) | LFV · KTH-TIPS2 / FMD / DTD | accuracy (%) | Hurt |
Closest papers by meaning
- Compositional Model Based Fisher Vector Coding for Image Classification (TPAMI 2017 · similarity 0.88)
- Deep TEN: Texture Encoding Network (CVPR 2017 · similarity 0.84)
- Semantic Fisher Scores for Task Transfer: Using Objects to Classify Scenes (TPAMI 2019 · similarity 0.84)
- Visualizing and Understanding Deep Texture Representations (CVPR 2016 · similarity 0.84)
- Analyzing Classifiers: Fisher Vectors and Deep Neural Networks (CVPR 2016 · similarity 0.84)
- Deep Filter Banks for Texture Recognition, Description, and Segmentation (IJCV 2016 · 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.
