目标检测发展进程:
YOLOv3复现代码合集涵盖 5 种常用深度学习框架:
Project
Infernece
Train
Star
gluoncv
√
√
3187
3.1使用YOLOv3训练、使用Mask-RCNN训练、理解ResNet、模型部署、人脸识别、文本分类等:
3.2基于yolo3 与crnn 实现中文自然场景文字检测及识别
3.3 YOLOv3 in PyTorch > ONNX > CoreML > iOS
3.4YoloV3/tiny-YoloV3+RaspberryPi3/Ubuntu LaptopPC+NCS/NCS2+USB Camera+Python+OpenVINO
3.5基于darknet框架实现CTPN版本自然场景文字检测与CNN+CTCOCR文字识别
3.6教程:用YOLO+Tesseract实现定制OCR系统
《Tutorial : Building a custom OCR using YOLO and Tesseract》
Traffic Signal Violation Detection System using Computer Vision - A Computer Vision based Traffic Signal Violation Detection System from video footage using YOLOv3 & Tkinter. (GUI Included)
3.8 OpenCV 'dnn' with NVIDIA GPUs: 1549% faster YOLO, SSD, and Mask R-CNN
3.9 Object Detection and Tracking
加入关键点的darknet训练框架,使用yolov3实现了轻量级的人脸检测
3.11 基于D/CIoU_YOLO_V3口罩识别
3.12 Object Detection: YOLO, MobileNetv3 and EfficientDet
3.13 Yolo-Fastest:超超超快的开源ARM实时目标检测算法
Network
VOC mAP(0.5)
COCO mAP(0.5)
Resolution
Run Time(Ncnn 1xCore)
Run Time(Ncnn 4xCore)
FLOPS
Weight size
MobileNetV2-YOLOv3-Nano
65.27
30.13
320
11.36ms
5.48ms
0.55BFlops
3.0MB
Yolo-Fastest(our)
61.02
23.65
320
6.74ms
4.42ms
0.23BFlops
1.3MB
Yolo-Fastest-XL(our)
69.43
32.45
320
15.15ms
7.09ms
0.70BFlops
3.5MB
3.15 OpenCV ‘dnn’ with NVIDIA GPUs: 1549% faster YOLO, SSD, and Mask R-CNN
在yolov5的基础上增加landmark预测分支,loss使用wingloss,使用yolov5s取得了相对于retinaface-r50更好的性能
5.1 Enriching Variety of Layer-wise Learning Information by Gradient Combination
Model
Size
mAP@0.5
BFLOPs
EfficientNet_b0-PRN
416x416
45.5
3.730
EfficientNet_b0-PRN
320x320
41.0
2.208
5.2 Gaussian YOLOv3: An Accurate and Fast Object Detector Using Localization Uncertainty for Autonomous Driving
5.3 YOLO Nano: a Highly Compact You Only Look Once Convolutional Neural Network for Object Detection
Model
model Size
mAP(voc 2007)
computational cost(ops)
Tiny YOLOv2[13]
60.5MB
57.1%
6.97B
Tiny YOLOv3[14]
33.4MB
58.4%
5.52B
YOLO Nano
4.0MB
69.1%
4.57B
5.4YOLO-LITE: A Real-Time Object Detection Algorithm Optimized for Non-GPU Computers
DataSet
mAP
FPS
PASCAL VOC
33.57
21
COCO
12.26
21
5.5 SlimYOLOv3: Narrower, Faster and Better for Real-Time UAV Applications
5.6 Strongeryolo-pytorch - Pytorch implementation of Stronger-Yolo with channel-pruning
Performance on VOC2007 Test(mAP) after pruning
Model
Backbone
MAP
Flops(G)
Params(M)
strongerv3
Mobilev2
79.6
4.33
6.775
strongerv3-sparsed
Mobilev2
77.4
4.33
6.775
strongerv3-Pruned(30% pruned)
Mobilev2
77.1
3.14
3.36
strongerv2
Darknet53
80.2
49.8
61.6
strongerv2-sparsed
Darknet53
78.1
49.8
61.6
strongerv2-Pruned(20% pruned)
Darknet53
76.8
49.8
45.2
5.7 Learning Spatial Fusion for Single-Shot Object Detection
YOLOv3+ASFF(自适应空间特征融合)组合,性能优于CornerNet和CenterNet等,在COCO上,38.1mAP/60 FPS,43.9mAP/29FPS!
System
test-dev mAP
Time (V100)
Time (2080ti)
YOLOv3 608
33.0
20ms
24ms
YOLOv3 608+ BoFs
37.0
20ms
24ms
YOLOv3 608(ours baseline)
38.8
20ms
24ms
YOLOv3 608+ ASFF
40.6
22ms
28ms
YOLOv3 608+ ASFF*
42.4
22ms
29ms
YOLOv3 800+ ASFF*
43.9
34ms
40ms
5.8 Distance-IoU Loss: Faster and Better Learning for Bounding Box Regression
5.9 xYOLO: A Model For Real-Time Object Detection In Humanoid Soccer On Low-End Hardware
5.10、CSPNet: A New Backbone that can Enhance Learning Capability of CNN
5.11、Spiking-YOLO: Spiking Neural Network for Real-time Object Detection
5.12、 Enriching Variety of Layer-wise Learning Information by Gradient Combination
5.13、YOLOv4: Optimal Speed and Accuracy of Object Detection
5.14、PP-YOLO: An Effective and Efficient Implementation of Object Detector
5.15、Scaled-YOLOv4: Scaling Cross Stage Partial Network
5.16、You Only Look One-level Feature
5.17、You Only Learn One Representation: Unified Network for Multiple Tasks
5.18、YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors
5.19、YOLOv6: A Single-Stage Object Detection Framework for Industrial Applications
5.20、YOLOX: Exceeding YOLO Series in 2021
5.21 You Only 👀 Once for Panoptic 🚗 Perception
5.22 YOLOPv2:rocket:: Better, Faster, Stronger for Panoptic driving Perception