Research map / Detection and segmentation / CNN features
Learning Data Augmentation Strategies for Object Detection
Barret Zoph, Ekin D. Cubuk, Golnaz Ghiasi, Tsung-Yi Lin, Jonathon Shlens, Quoc V. Le
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 learned augmentation policy | RetinaNet ResNet-50 · COCO | mAP: 36.7 -> 39.0 | Improved |
| add learned augmentation policy | RetinaNet ResNet-101 · COCO | mAP: 38.8 -> 40.4 | Improved |
| add learned augmentation policy | RetinaNet ResNet-200 · COCO | mAP: 39.9 -> 42.1 | Improved |
| add geometric operations to color operations in the policy (ablation) | learned policy · COCO ResNet-50 | mAP: 37.5 -> 38.6 | Improved |
| add bbox-only operations to the policy (ablation) | learned policy · COCO ResNet-50 | mAP: 38.6 -> 39.0 | Improved |
| use learned augmentation instead of DropBlock regularization | RetinaNet ResNet-50 · COCO | mAP: 38.4 -> 39.0 | Improved |
| add learned augmentation | AmoebaNet NAS-FPN · COCO val | mAP: 47.0 -> 48.6 | Improved |
| increase image size to 1536 and anchors from 3x3 to 9x9 (ablation) | AmoebaNet NAS-FPN with learned augmentation · COCO val | mAP: 48.6 -> 50.7 | Improved |
| add policy learned on COCO | Faster R-CNN ResNet-101 · PASCAL VOC 2007 | mAP50: 76.0 -> 78.7 | Improved |
| add learned augmentation with only 5000 training images | RetinaNet ResNet-50 · COCO subset 5000 | mAP: 6.5 -> 8.7 | Improved |
| add learned augmentation with 23000 training images | RetinaNet ResNet-50 · COCO subset 23000 | mAP: 22.6 -> 25.3 | Improved |
| add learned augmentation, small objects | RetinaNet ResNet-50 · COCO subset 5000 | mAP small: 1.9 -> 3.2 | Improved |
Closest papers by meaning
- Robust and Accurate Object Detection via Adversarial Learning (CVPR 2021 · similarity 0.91)
- Scale-Aware Automatic Augmentations for Object Detection With Dynamic Training (TPAMI 2022 · similarity 0.90)
- Modeling Visual Context is Key to Augmenting Object Detection Datasets (ECCV 2018 · similarity 0.90)
- A Simple Background Augmentation Method for Object Detection with Diffusion Model (ECCV 2024 · similarity 0.89)
- RandAugment: Practical Automated Data Augmentation with a Reduced Search Space (NeurIPS 2020 · similarity 0.89)
- The good, the bad and the ugly sides of data augmentation: An implicit spectral regularization perspective (JMLR 2024 · 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.
