Research map / Detection and segmentation / Weakly supervised detection and segmentation
Tobias: A Random CNN Sees Objects
Yun-Hao Cao, Jianxin Wu
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 |
|---|---|---|---|
| add ReLU instead of sigmoid (ablation) | SCDA random R-50 · CUB-200 localization | accuracy: 22.6 -> 41.8 | Improved |
| add network depth: bottlenecks [1,1,1,1] to [3,4,6,3] (ablation) | SCDA random R-50 · ImageNet localization | accuracy: 42.1 -> 48.2 | Improved |
| make network deeper: [3,4,6,3] to [6,8,12,6] (ablation) | SCDA random R-50 · ImageNet localization | accuracy: 48.2 -> 50.0 | Improved |
| add skip connections (ablation) | random R-50 · ImageNet localization | accuracy: 50.8 -> 48.2 | No clear effect |
| add batch normalization (ablation) | random R-50 · ImageNet localization | accuracy: 49.1 -> 48.2 | No clear effect |
| use Kaiming instead of Xavier initialization (ablation) | random R-50 · ImageNet localization | accuracy: 42.2 -> 48.2 | Improved |
| use CNN instead of ViT-Base for random-init localization (ablation) | SCDA · ImageNet localization | accuracy: 40.5 -> 48.2 | Improved |
| add Tobias background-swap augmentation (ablation) | MoCov2 ResNet-50 CUB-200 200 ep · CUB-200 fine-tuning | accuracy: 56.2 -> 63.6 | Improved |
| add Tobias augmentation (ablation) | MoCov2 ResNet-50 CUB-200 800 ep · CUB-200 fine-tuning | accuracy: 66.5 -> 67.2 | Improved |
| add Tobias augmentation (ablation) | MoCov2 IN-10k 800 ep · VOC07+12 detection | AP75: 45.9 -> 48.5 | Improved |
| add Tobias augmentation (ablation) | MoCov2 IN-10k 200 ep · Tiny-IN-200 linear | accuracy: 0.5 -> 9.9 | Improved |
| add Tobias augmentation (ablation) | MoCov2 IN-50k 200 ep · Tiny-IN-200 linear | accuracy: 26.3 -> 26.0 | Hurt |
Closest papers by meaning
- A Random CNN Sees Objects: One Inductive Bias of CNN and Its Applications (AAAI 2022 · similarity 0.98)
- Spatially Invariant Unsupervised Object Detection with Convolutional Neural Networks (AAAI 2019 · similarity 0.86)
- Unsupervised Object Localization: Observing the Background To Discover Objects (CVPR 2023 · similarity 0.86)
- Finding Distributed Object-Centric Properties in Self-Supervised Transformers (CVPR 2026 · similarity 0.86)
- Grounding Everything: Emerging Localization Properties in Vision-Language Transformers (CVPR 2024 · similarity 0.84)
- Believing is Seeing: Unobserved Object Detection using Generative Models (CVPR 2025 · 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.
