Research map / Architectures and efficiency / CNN design
Factors of Transferability for a Generic ConvNet Representation
Hossein Azizpour, Ali Sharif Razavian, Josephine Sullivan, Atsuto Maki, Stefan Carlsson
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 |
|---|---|---|---|
| optimize all transferability factors vs standard FC6 representation | Medium AlexNet ConvNet + linear SVM · VOC07 | accuracy (%): 71.8 -> 80.7 | Improved |
| optimize all transferability factors vs standard FC6 representation | Medium AlexNet ConvNet · Pet | accuracy (%): 78.5 -> 88.1 | Improved |
| optimize all transferability factors vs standard FC6 representation | Medium AlexNet ConvNet · UKB | score: 93.0 -> 96.3 | Improved |
| optimize factors vs standard (non-ConvNet state of the art stays better) | Deep Optimized · Oxford buildings retrieval | mAP (%): 81.7 non-ConvNet vs 79.0 optimized | Mixed |
| add fine-tuning to the ConvNet (ablation) | Medium FC7 · CUB | accuracy (%): 62.9 -> 66.4 | Improved |
| add fine-tuning to the ConvNet (ablation) | Medium FC7 · MIT | accuracy (%): 65.9 -> 66.3 | Improved |
| add fine-tuning to the ConvNet (ablation) | Medium FC7 · Flower | accuracy (%): 90.4 -> 91.4 | Improved |
| train on Places scene labels instead of ImageNet (ablation) | AlexNet-style ConvNet · Pet | accuracy (%): 78.4 -> 49.9 | Hurt |
| train on Places scene labels instead of ImageNet (ablation) | AlexNet-style ConvNet · MIT | accuracy (%): 64.9 -> 69.3 | Improved |
| concatenate ImageNet and Places features (ablation) | ImageNet ConvNet · VOC07 | accuracy (%): 71.6 -> 73.8 | Improved |
| train on more source images (SUN397 130K to Places 2.5M) (ablation) | scene ConvNet · MIT | accuracy (%): 62.6 -> 69.3 | Improved |
| fine-tune with extra Pet and CUB images in addition to VOC (ablation) | ConvNet detector · VOC07 cat | mAP (%): 60.7 -> 63.0 | Improved |
Closest papers by meaning
- A ConvNet for the 2020s (CVPR 2022 · similarity 0.88)
- Analyzing Filters Toward Efficient ConvNet (CVPR 2018 · similarity 0.88)
- ConvNeXt V2: Co-Designing and Scaling ConvNets With Masked Autoencoders (CVPR 2023 · similarity 0.87)
- Conv2Former: A Simple Transformer-Style ConvNet for Visual Recognition (TPAMI 2024 · similarity 0.86)
- ConvNet vs Transformer, Supervised vs CLIP: Beyond ImageNet Accuracy (ICML 2024 · similarity 0.86)
- Improving the Adversarial Transferability of Vision Transformers with Virtual Dense Connection (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.
