Research map / Detection and segmentation / Weakly supervised detection and segmentation
Weakly Supervised Complementary Parts Models for Fine-Grained Image Classification From the Bottom Up
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Add complementary parts with stacked LSTM and multi-loss fusion | GoogleNet · Stanford Dogs 120 | accuracy: 85.2 -> 93.9 | Helped |
| Use multi-loss instead of single loss (ablation) | Stacked LSTM complementary-parts model · Stanford Dogs 120 | accuracy: 92.4 -> 93.9 | Helped |
| Add complementary parts with stacked LSTM and multi-loss fusion | GoogleNet · CUB200 | accuracy: 82.6 -> 90.3 | Helped |
| SJFT with ImageNet images | GoogleNet complementary-parts method · CUB200 | accuracy: 90.3 -> 90.4 | No clear effect |
| Add complementary parts with stacked LSTM and multi-loss fusion | GoogleNet · Caltech 256 | accuracy: 84.1±0.2 -> 93.5±0.2 | Helped |
| Use multi-loss instead of single loss (ablation) | Stacked LSTM complementary-parts model · CUB200 | accuracy: 87.6 -> 90.3 | Helped |
| Use multi-loss instead of single loss (ablation) | Stacked LSTM complementary-parts model · Caltech 256 | accuracy: 90.1±0.2 -> 93.5±0.2 | Helped |
| Vary the number of complementary parts; n=9 performs best among tested counts (ablation) | Complementary parts model · CUB200 | accuracy: 85.3%, 87.9%, 89.1%, 90.3%, 87.6%, 86.8%, 85.9% | Mixed |
| Use object features only rather than the reference model (ablation) | Complementary parts model · CUB200 | accuracy: 90.3% -> 90.0% | Hurt |
| Use image features only rather than the reference model (ablation) | Complementary parts model · CUB200 | accuracy: 90.3% -> 82.8% | Hurt |
| Use unoptimized uniform-grid cells as object parts (ablation) | Complementary parts model · CUB200 | accuracy: 90.3% -> 78.3% | Hurt |
| Randomly sample n=9 suppressed proposals instead of grid-based initialization (ablation) | Complementary parts model · CUB200 | accuracy: 90.3% -> 86.9% | Hurt |
Closest papers by meaning
- Part-Stacked CNN for Fine-Grained Visual Categorization (CVPR 2016 · similarity 0.89)
- Weakly Supervised Learning of Part Selection Model with Spatial Constraints for Fine-Grained Image Classification (AAAI 2017 · similarity 0.89)
- Combining Bottom-Up, Top-Down, and Smoothness Cues for Weakly Supervised Image Segmentation (CVPR 2017 · similarity 0.88)
- SPDA-CNN: Unifying Semantic Part Detection and Abstraction for Fine-Grained Recognition (CVPR 2016 · similarity 0.88)
- Unsupervised Part Learning for Visual Recognition (CVPR 2017 · similarity 0.88)
- Weakly Supervised Deep Detection Networks (CVPR 2016 · similarity 0.88)
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.