Simple Training Strategies and Model Scaling for Object Detection

Abstract

The speed-accuracy Pareto curve of object detection systems have advanced through a combination of better model architectures, training and inference methods. In this paper, we methodically evaluate a variety of these techniques to understand where most of the improvements in modern detection systems come from. We benchmark these improvements on the vanilla ResNet-FPN backbone with RetinaNet and RCNN detectors. The vanilla detectors are improved by 7.7% in accuracy while being 30% faster in speed. We further provide simple scaling strategies to generate family of models that form two Pareto curves, named RetinaNet-RS and Cascade RCNN-RS. These simple rescaled detectors explore the speed-accuracy trade-off between the one-stage RetinaNet detectors and two-stage RCNN detectors. Our largest Cascade RCNN-RS models achieve 52.9% AP with a ResNet152-FPN backbone and 53.6% with a SpineNet143L backbone. Finally, we show the ResNet architecture, with three minor architectural changes, outperforms EfficientNet as the backbone for object detection and instance segmentation systems.

8 pages

References in corpus (8)

• Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks
• YOLOv4: Optimal Speed and Accuracy of Object Detection
• YOLOv3: An Incremental Improvement
• Gaussian Error Linear Units (GELUs)
• Deep High-Resolution Representation Learning for Visual Recognition
• Scalability in Perception for Autonomous Driving: Waymo Open Dataset
• Hybrid Task Cascade for Instance Segmentation
• Efficient Scale-Permuted Backbone with Learned Resource Distribution

Cited by in corpus (3)

• Masked-attention Mask Transformer for Universal Image Segmentation
• Revisiting 3D ResNets for Video Recognition
• HAISTA-NET: Human Assisted Instance Segmentation Through Attention