Research map / Detection and segmentation / CNN features
Proper Reuse of Image Classification Features Improves Object Detection
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 |
|---|---|---|---|
| Freeze the pretrained backbone rather than fine-tune it | EfficientNet-B7 + NAS-FPN on MSCOCO · MSCOCO val | Box mAP: +0.9 Box mAP | Helped |
| Freeze the pretrained backbone rather than fine-tune it | EfficientNet-B7 + NAS-FPN on MSCOCO · MSCOCO val | Mask mAP: +0.7 Mask mAP | Helped |
| Add Copy-Paste augmentation and freeze the pretrained backbone (ablation) | EfficientNet-B7 + NAS-FPN · MSCOCO val | Box mAP: 55.9 -> 57.0 | Helped |
| Freeze the pretrained backbone | EfficientNet-B7 + NAS-FPN on LVIS · LVIS val | Box mAP: 37.2 -> 39.4 | Helped |
| Add Copy-Paste augmentation and freeze the pretrained backbone (ablation) | EfficientNet-B7 + NAS-FPN · LVIS val | Box mAP: 41.6 -> 43.1 | Helped |
| Freeze the pretrained backbone | ResNet-101 + NAS-FPN + Cascade · MSCOCO, 600 epochs | mAP: 51.1 -> 51.8 | Helped |
| Freeze the pretrained backbone (ablation) | ResNet-101 + FPN · MSCOCO, 600 epochs | mAP: 48.6 -> 42.1 | Hurt |
| Use a higher-capacity NAS-FPN detector rather than FPN (ablation) | Frozen ResNet-101 detector · MSCOCO | mAP: -6.5 -> +0.1 | Helped |
| Pretrain on JFT-300M rather than ImageNet while freezing the backbone (ablation) | ResNet detectors with NAS-FPN and Cascade · MSCOCO | mAP: +0.9 mAP in most cases | Helped |
| Add residual adapters to the frozen backbone (ablation) | ResNet-101 + NAS-FPN + Cascade · MSCOCO | mAP: 51.8 -> 53.0 | Helped |
| Increase detector hidden representation size from 256 to 512 (ablation) | Frozen ResNet-101 + FPN · MSCOCO, 72 epochs | mAP: 40.9 -> 42.9 | Helped |
| Reduce RPN convolutional layers from two to one (ablation) | Frozen ResNet-101 + FPN · MSCOCO | mAP: 42.1 -> 41.3 | Hurt |
Closest papers by meaning
- Distilling Image Classifiers in Object Detectors (NeurIPS 2021 · similarity 0.88)
- Frozen-DETR: Enhancing DETR with Image Understanding from Frozen Foundation Models (NeurIPS 2024 · similarity 0.87)
- Revisiting Knowledge Transfer for Training Object Class Detectors (CVPR 2018 · similarity 0.87)
- Feature Intertwiner for Object Detection (ICLR 2019 · similarity 0.87)
- Instance-Aware, Context-Focused, and Memory-Efficient Weakly Supervised Object Detection (CVPR 2020 · similarity 0.87)
- Long-tailed Object Detection Pretraining: Dynamic Rebalancing Contrastive Learning with Dual Reconstruction (NeurIPS 2024 · 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.