YOLOv4: Optimal Speed and Accuracy of Object Detection
By Alexey Bochkovskiy, Chien-Yao Wang, Hongpeng Liao. 2020
15,853 citations recorded by the source corpus.
There are a huge number of features which are said to improve Convolutional Neural Network (CNN) accuracy. Practical testing of combinations of such features on large datasets, and theoretical justification of the result, is required. Some features operate on certain models exclusively and for certain problems exclusively, or only for small-scale datasets; while some features, such as batch-normalization and residual-connections, are applicable to the majority of models, tasks, and datasets. We assume that such universal features include Weighted-Residual-Connections (WRC), Cross-Stage-Partial-connections (CSP), Cross mini-Batch Normalization (CmBN), Self-adversarial-training (SAT) and Mish-activation. We use new features: WRC, CSP, CmBN, SAT, Mish activation, Mosaic data augmentation, CmBN, DropBlock regularization, and CIoU loss, and combine some of them to achieve state-of-the-art results: 43.5% AP (65.7% AP50) for the MS COCO dataset at a realtime speed of ~65 FPS on Tesla V100. Source code is at this https URL
Subfields: Computer Vision and Pattern Recognition, Electrical and Electronic Engineering, Industrial and Manufacturing Engineering
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Computer Vision and Pattern Recognition
- Luc Van Gool
- Dacheng Tao
- Ming-Hsuan Yang
- Xiaogang Wang
- Trevor Darrell
- Jitendra Malik
- Bernt Schiele
- Chunhua Shen
In Electrical and Electronic Engineering
In Industrial and Manufacturing Engineering
- Diganta Misra
- Hongpeng Liao
- Chien-Yao Wang
- Jian‐Yuan Sun
- Upesh Nepal
- D. Bull
- Nantheera Anantrasirichai
- Yong Xu
Citations in the corpus
- Inner Monologue: Embodied Reasoning through Planning with Language Models (2022, background): “One common type of such feedback is object recognition [89, 90, 91, 92] that returns a list of”
- Simple Copy-Paste is a Strong Data Augmentation Method for Instance Segmentation (2020, methodology): “The Mosaic data augmentation method employed in YOLOv4 [1] is related to CutMix in the sense that one creates a new compound image that is a rectangular grid of multiple individual images along with their ground truths.”
- DAB-DETR: Dynamic Anchor Boxes are Better Queries for DETR (2022, background): “Most classical detectors are based on convolutional architectures which have made remarkable progress in the last decade (Ren et al., 2017; Girshick, 2015; Redmon et al., 2016; Bochkovskiy et al., 2020; Ge et al., 2021).”
- UAV-YOLOv8: A Small-Object-Detection Model Based on Improved YOLOv8 for UAV Aerial Photography Scenarios (2023, methodology): “The YOLO family of algorithms used in this paper are as follows: YOLOv3 [10], which uses multiscale detection for the first time, and its lightweight version, YOLOv3-tiny; YOLOv4 [11] uses the idea of CSPNet [48] to construct a new backbone network structure called CSPDarknet53, which is lightweight ”
- Medical Image Segmentation Using Deep Learning: A Survey (2020, methodology): “In this work, the first step is to use the regional deep learning method YOLO for target detection, the second step is to input the detected targets into a newly designed full-resolution convolutional network (FrCN) for segmentation, and finally, a deep convolutional neural network is used to identi”
- Learn to Predict Sets Using Feed-Forward Neural Networks (2020, background): “grid [20], [21], [22], [23] or anchor boxes [24] representing a coarse estimate of all possible locations and scales of objects.”
- Deep Learning-based Human Pose Estimation: A Survey (2020, methodology): “However, with real-time object detectors such as YOLOv4 [287], the processing speed of top-down approaches can be greatly improved.”
- Artificial intelligence in the creative industries: a review (2020, methodology): “rates (currently reported up to 55 fps). YOLO divides an image into regions and predicts bounding boxes using a multi-scale approach and gives probabilities for each region. The latest model, YOLOv4, [183], concatenates YOLOv3 [172] with a convolutional neural network that is 53 layers deep, with SP”
- TPH-YOLOv5: Improved YOLOv5 Based on Transformer Prediction Head for Object Detection on Drone-captured Scenarios (2021, methodology): “Generally, YOLOv5 respectively uses the architecture of CSPDarknet53 with an SPP layer as backbone, PANet as Neck and YOLO detection head [37].”
- DeFRCN: Decoupled Faster R-CNN for Few-Shot Object Detection (2021, background): “, twostage proposal-based paradigm [4, 8, 14, 15, 16, 23, 36] and one-stage proposal-free one [3, 25, 33, 34, 35], which both have witnessed fantastic progress on numerous large-scale benchmarks.”
- Activation functions in deep learning: A comprehensive survey and benchmark (2021, methodology): “It has recently been used by the YOLOv4 model for object detection [100].”
- YOLO-based Object Detection Models: A Review and its Applications (2024, neutral): “10 [76], used the concatenation method which improves the overall prediction accuracy.”
- MPDIoU: A Loss for Efficient and Accurate Bounding Box Regression (2023, methodology): “Most of the state-of-the-art object detectors (e.g., YOLO series [1, 2, 3, 4, 5, 6], Mask R-CNN [7], Dynamic R-CNN [8] and DETR [9]) rely on a bounding box regression (BBR) module to determine the position of objects.”
- Object detection using YOLO: challenges, architectural successors, datasets and applications (2022, background): “The fourth version of YOLO series was drafted by Alexey Bochkovskiy, Chien-Yao Wang and Hong-Yuan Mark Liao [5].”