Research map / Detection and segmentation / CNN features
Scene Recognition With CNNs: Objects, Scales and Dataset Bias
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 |
|---|---|---|---|
| Assign Places-CNN to global scales and ImageNet-CNN to local scales | Multi-scale scene recognition · SUN397 | accuracy: 63.81 -> 65.38 | Helped |
| Assign Places-CNN to global scales and ImageNet-CNN to local scales | Multi-scale scene recognition · MIT Indoor 67 | accuracy: 78.21 -> 80.97 | Helped |
| Combine Places-CNN global features with ImageNet-CNN local features | Two-network scene recognition · SUN397 | accuracy: 58.87 -> 64.10 | Helped |
| Combine Places-CNN global features with ImageNet-CNN local features | Two-network scene recognition · MIT Indoor 67 | accuracy: 74.18 -> 78.28 | Helped |
| Combine full ImageNet-CNN and Places-CNN features across seven scales | Multi-scale AlexNet scene recognition · SUN397 | accuracy: 65.38 -> 66.26 | Helped |
| Add fine tuning to the pretrained networks (ablation) | Single-scale recognition on MIT Indoor 67 · MIT Indoor 67 | accuracy: 61.49 (68.21); 74.18 (73.35) | Mixed |
| Use VGG rather than AlexNet in the three-scale hybrid (ablation) | Three-scale hybrid scene recognition · SUN397 | accuracy: 63.03 -> 70.17 | Helped |
| Use VGG rather than AlexNet in the three-scale hybrid (ablation) | Three-scale hybrid scene recognition · MIT Indoor 67 | accuracy: 78.28 -> 86.04 | Helped |
| Use Places-CNN at all scales instead of a scale-specific hybrid (ablation) | Seven-scale scene recognition · SUN397 | accuracy: 65.38 -> 63.81 | Hurt |
| Use ImageNet-CNN at all scales instead of a scale-specific hybrid (ablation) | Seven-scale scene recognition · SUN397 | accuracy: 65.38 -> 58.71 | Hurt |
| Use ImageNet-CNN at all scales instead of a scale-specific hybrid (ablation) | Seven-scale scene recognition · MIT Indoor 67 | accuracy: 80.97 -> 74.33 | Hurt |
| Use a single combined ImageNet+Places training set in Hybrid-CNN | Scene recognition · SUN397 | accuracy: 58.87 -> 53.86 | Hurt |
Closest papers by meaning
- ScaleNet - Improve CNNs through Recursively Rescaling Objects (AAAI 2020 · similarity 0.88)
- ScaleNet: A Shallow Architecture for Scale Estimation (CVPR 2022 · similarity 0.88)
- Data-Driven Neuron Allocation for Scale Aggregation Networks (CVPR 2019 · similarity 0.88)
- An Empirical Evaluation of Current Convolutional Architectures' Ability to Manage Nuisance Location and Scale Variability (CVPR 2016 · similarity 0.87)
- Depth CNNs for RGB-D Scene Recognition: Learning from Scratch Better than Transferring from RGB-CNNs (AAAI 2017 · similarity 0.87)
- Attention to Scale: Scale-Aware Semantic Image Segmentation (CVPR 2016 · similarity 0.87)
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.