Research map / Label-efficient and robust learning / Self-supervised and contrastive
Learning Representations by Predicting Bags of Visual Words
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 |
|---|---|---|---|
| Train BoWNet to predict visual-word BoW targets from perturbed images | RotNet representations · CIFAR-100 | linear accuracy: 60.3 -> 71.5 | Helped |
| Train BoWNet to predict visual-word BoW targets from perturbed images | RotNet representations · VOC07+12 detection on test07 | AP all: 53.2 -> 55.8 | Helped |
| Apply BoWNet iteratively for three training rounds (ablation) | BoWNet · CIFAR-100 | linear accuracy: 71.5 -> 74.5 | Helped |
| Remove CutMix augmentation (ablation) | BoWNet · CIFAR-100 | linear accuracy: 71.5 -> 69.7 | Hurt |
| Add spatial-pyramid BoW targets (ablation) | BoWNet · CIFAR-100 | linear accuracy: 71.5 -> 71.3 | No clear effect |
| Use the perturbed image's BoW as the prediction target instead of the original image's BoW (ablation) | BoWNet · CIFAR-100 | linear accuracy: 71.5 -> 61.3 | Hurt |
| Replace the reparameterized prediction layer with a standard linear layer (ablation) | BoWNet · CIFAR-100 | linear accuracy: 71.5 -> 63.3 | Hurt |
| Use binary rather than histogram BoW targets (ablation) | BoWNet · CIFAR-100 | linear accuracy: 71.5 -> 71.4 | No clear effect |
| Increase vocabulary size from K=2048 to K=4096 (ablation) | BoWNet · CIFAR-100 | linear accuracy: 71.01 -> 70.99 | No clear effect |
| Use RelLoc rather than RotNet to build BoW targets (ablation) | BoWNet · MiniImageNet novel classes 10-shot | accuracy: 74.0 -> 75.9 | Helped |
| Build BoW targets from a randomly initialized base convnet (ablation) | BoWNet · MiniImageNet novel classes 10-shot | accuracy: 74.0 -> 68.9 | Hurt |
| Build BoW targets from conv5 rather than conv4 (ablation) | BoWNet · ImageNet small-scale | linear accuracy: 45.38 -> 40.37 | Hurt |
Closest papers by meaning
- Revisiting Self-Supervised Visual Representation Learning (CVPR 2019 · similarity 0.90)
- Self-Supervised Learning of Visual Features Through Embedding Images Into Text Topic Spaces (CVPR 2017 · similarity 0.89)
- Self-Supervised Learning as Discrete Communication (ICML 2026 · similarity 0.89)
- A Vision Check-up for Language Models (CVPR 2024 · similarity 0.87)
- Self-Organizing Visual Prototypes for Non-Parametric Representation Learning (ICML 2025 · similarity 0.86)
- TIPS: Text-Image Pretraining with Spatial awareness (ICLR 2025 · similarity 0.86)
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.