Research map / Detection and segmentation / CNN features
MegDet: A Large Mini-Batch Object Detector
Read the paperCheck 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 |
|---|---|---|---|
| Increase batch size from 16 to 32 without BN | ResNet-50 + FPN detector · COCO val | mmAP: 36.2 -> 36.4 | Helped |
| Increase batch size from 16 to 32 without BN | ResNet-50 + FPN detector · COCO val | training time (h): 33.2 -> 15.1 | Helped |
| Add warmup and Cross-GPU Batch Normalization | Large-mini-batch ResNet-50 + FPN detector · COCO val | mmAP: 36.2 -> 37.1 | Helped |
| Add warmup and Cross-GPU Batch Normalization | Large-mini-batch ResNet-50 + FPN detector · COCO val | convergence | Helped |
| Increase BN statistics size from 2 to 16 images (ablation) | Batch-16 ResNet-50 + FPN detector · COCO val | mmAP: 31.0 -> 37.0 | Helped |
| Increase BN statistics size from 16 to 64 images (ablation) | Batch-64 ResNet-50 + FPN detector · COCO val | mmAP: 37.1 -> 35.3 | Hurt |
| Use the long training policy instead of the normal policy (ablation) | Batch-32 detector with BN size 32 · COCO val | mmAP: 37.3 -> 37.8 | Helped |
| Increase batch size from 16 to 256 with CGBN | ResNet-50 + FPN detector · COCO val | training time (h): 33.2 -> 4.1 | Helped |
| Use batch size 64 without BN at the scaled learning rate (ablation) | ResNet-50 + FPN detector · COCO val | convergence | Hurt |
| Use batch size 128 with half learning rate and no BN (ablation) | ResNet-50 + FPN detector · COCO val | convergence | Hurt |
Closest papers by meaning
- ScratchDet: Training Single-Shot Object Detectors From Scratch (CVPR 2019 · similarity 0.88)
- QueryDet: Cascaded Sparse Query for Accelerating High-Resolution Small Object Detection (CVPR 2022 · similarity 0.87)
- MobileDets: Searching for Object Detection Architectures for Mobile Accelerators (CVPR 2021 · similarity 0.87)
- DynamicDet: A Unified Dynamic Architecture for Object Detection (CVPR 2023 · similarity 0.87)
- M2Det: A Single-Shot Object Detector Based on Multi-Level Feature Pyramid Network (AAAI 2019 · similarity 0.87)
- Unsupervised Object Detection with Theoretical Guarantees (NeurIPS 2024 · 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.