Bag of Freebies for Training Object Detection Neural Networks

By Zhi-Li Zhang, Tong He, Hang Zhang, Zhongyue Zhang, Junyuan Xie, Mu Li. 2019

203 citations recorded by the source corpus.

Training heuristics greatly improve various image classification model accuracies~\cite{he2018bag}. Object detection models, however, have more complex neural network structures and optimization targets. The training strategies and pipelines dramatically vary among different models. In this works, we explore training tweaks that apply to various models including Faster R-CNN and YOLOv3. These tweaks do not change the model architectures, therefore, the inference costs remain the same. Our empirical results demonstrate that, however, these freebies can improve up to 5% absolute precision compared to state-of-the-art baselines.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Optimization & Training

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Quoc V. Le
2. Tsung-Yi Lin
3. Barret Zoph
4. Ekin Dogus Cubuk
5. Alan L. Yuille
6. Jianfeng Gao
7. Golnaz Ghiasi
8. Yin Cui

In Computer Vision and Pattern Recognition

1. Luc Van Gool
2. Xiaogang Wang
3. Alan L. Yuille
4. Dacheng Tao
5. Ming-Hsuan Yang
6. Trevor Darrell
7. Andrew Zisserman
8. Jitendra Malik

In Optimization & Training

1. Quoc V. Le
2. Ekin Dogus Cubuk
3. Barret Zoph
4. Tsung-Yi Lin
5. Golnaz Ghiasi
6. A. Srinivas
7. Zizhao Zhang
8. Kihyuk Sohn

Citations in the corpus

• Simple Copy-Paste is a Strong Data Augmentation Method for Instance Segmentation (2020, methodology): “On the other hand, as it is shown in Figure 5, mixup [66, 69] data augmentation does not help when it is used with large scale jittering.”
• AdaptIS: Adaptive Instance Selection Network (2019, methodology): “We use MXNet Gluon [3] with GluonCV [29] framework for training and inference of our models.”
• RTMDet: An Empirical Study of Designing Real-Time Object Detectors (2022, background): “Furthermore, inhibiting weight decay on normalization layers and biases following previous practices [96] brings 0.”
• A Simple Semi-Supervised Learning Framework for Object Detection (2020, methodology): “Some works have presented augmentation techniques for supervised object detection, such as MixUp [64,65], CutMix [61], or augmentation strategy learning [69].”
• Learning Spatial Fusion for Single-Shot Object Detection (2019, methodology): “To make fair comparison, we further report the accuracies of another two common YOLOv3 @608 BoF [43] GA [38] IoU [41] fusion operations ( i.e. element-wise sum and concatenation) with resized features in Table 3.”
• Comparing YOLOv3, YOLOv4 and YOLOv5 for Autonomous Landing Spot Detection in Faulty UAVs (2022, methodology): “Finally, YOLO layer generates the results, similar to YOLOv3 YOLOv4 uses bag of freebies [34] and bag of specials [7] to improve the algorithm performance.”
• YOLOX: Exceeding YOLO Series in 2021 (2021, background): “MixUp [10] is originally designed for image classification task but then modified in BoF [38] for object detection training.”