BestYOLO：以科研和竞赛为导向的最好的YOLO实践框架!

<div align="center"><img src="./images/home1.png" /></div> BestYOLO是一个以科研和竞赛为导向的最好的YOLO实践框架! 目前BestYOLO是一个完全基于[YOLOv5 v7.0](https://github.com/ultralytics/yolov5/tree/v7.0) 进行改进的开源库，该库将始终秉持以落地应用为导向，以轻便化使用为宗旨，简化各种模块的改进。目前已经集成了基于[torchvision.models](https://pytorch.org/vision/stable/index.html) 模型为Backbone的YOLOv5目标检测算法，同时也将逐渐开源更多YOLOv5应用程序。 # 🌟改进 - [Backbone-ResNet18](https://github.com/WangRongsheng/BestYOLO/blob/main/models/backbone/resnet18.yaml) 对齐 [resnet18](https://pytorch.org/vision/stable/models/generated/torchvision.models.resnet18.html#torchvision.models.resnet18) - [Backbone-RegNet_y_400mf](https://github.com/WangRongsheng/BestYOLO/blob/main/models/backbone/RegNety400.yaml) 对齐 [regnet_y_400mf](https://pytorch.org/vision/stable/models/generated/torchvision.models.regnet_y_400mf.html#torchvision.models.regnet_y_400mf) - [Backbone-MobileNetV3 small](https://github.com/WangRongsheng/BestYOLO/blob/main/models/backbone/MobileNetV3s.yaml) 对齐 [mobilenet_v3_small](https://pytorch.org/vision/stable/models/generated/torchvision.models.mobilenet_v3_small.html#torchvision.models.mobilenet_v3_small) - [Backbone-EfficientNet_B0](https://github.com/WangRongsheng/BestYOLO/blob/main/models/backbone/efficientnet_b0.yaml) 对齐 [efficientnet_b0](https://pytorch.org/vision/stable/models/generated/torchvision.models.efficientnet_b0.html#torchvision.models.efficientnet_b0) - [Backbone-ResNet34](https://github.com/WangRongsheng/BestYOLO/blob/main/models/backbone/resnet34.yaml) 对齐 [resnet34](https://pytorch.org/vision/stable/models/generated/torchvision.models.resnet34.html#torchvision.models.resnet34) - [Backbone-ResNet50](https://github.com/WangRongsheng/BestYOLO/blob/main/models/backbone/resnet50.yaml) 对齐 [resnet50](https://pytorch.org/vision/stable/models/generated/torchvision.models.resnet50.html#torchvision.models.resnet50) - [Backbone-EfficientNetV2_s](https://github.com/WangRongsheng/BestYOLO/blob/main/models/backbone/efficientnet_v2_s.yaml) 对齐 [efficientnet_v2_s](https://pytorch.org/vision/stable/models/generated/torchvision.models.efficientnet_v2_s.html#torchvision.models.efficientnet_v2_s) - [Backbone-EfficientNet_B1](https://github.com/WangRongsheng/BestYOLO/blob/main/models/backbone/efficientnet_b1.yaml) 对齐 [efficientnet_b1](https://pytorch.org/vision/stable/models/generated/torchvision.models.efficientnet_b1.html#torchvision.models.efficientnet_b1) - [Backbone-MobileNetV2](https://github.com/WangRongsheng/BestYOLO/blob/main/models/backbone/mobilenet_v2.yaml) 对齐 [mobilenet_v2](https://pytorch.org/vision/stable/models/generated/torchvision.models.mobilenet_v2.html#torchvision.models.mobilenet_v2) - [Backbone-wide_resnet50_2](https://github.com/WangRongsheng/BestYOLO/blob/main/models/backbone/wide_resnet50_2.yaml) 对齐 [wide_resnet50_2](https://pytorch.org/vision/stable/models/generated/torchvision.models.wide_resnet50_2.html#torchvision.models.wide_resnet50_2) - [Backbone-VGG11_BN](https://github.com/WangRongsheng/BestYOLO/blob/main/models/backbone/vgg11_bn.yaml) 对齐 [vgg11_bn](https://pytorch.org/vision/stable/models/generated/torchvision.models.vgg11_bn.html#torchvision.models.vgg11_bn) - [Backbone-Convnext Tiny](https://github.com/WangRongsheng/BestYOLO/blob/main/models/backbone/convnext_tiny.yaml) 对齐 [convnext_tiny](https://pytorch.org/vision/stable/models/generated/torchvision.models.convnext_tiny.html#torchvision.models.convnext_tiny) > 所有Backbone都支持开启预训练权重，只需添加`pretrained=True`到每个[common.py](https://github.com/WangRongsheng/BestYOLO/blob/main/models/common.py#L870) 的模型中。`torchvision.models`中的预训练权重都是基于ImageNet-1K数据集训练的！ |models|layers|parameters|model size(MB)| |:-|:-|:-|:-| |yolov5n|214|1766623|3.9| |MobileNetV3s|313|2137311|4.7| |efficientnet_b0|443|6241531|13.0| |RegNety400|450|5000191|10.5| |ResNet18|177|12352447|25.1| |ResNet34|223|22460607|45.3| |ResNet50|258|27560895|55.7| |EfficientNetV2_s|820|22419151|45.8| |efficientnet_b1|539|6595615|13.8| |mobilenet_v2|320|4455295|9.4| |wide_resnet50_2|258|70887103|142.3| |vgg11_bn|140|10442879|21.9| |convnext_tiny|308|29310175|59.0| > `.yaml`配置文件中的`depth_multiple`和`width_multiple`可以同时设置为1试试，说不定会有不错的效果。 > SPP是空间金字塔池化，作用是一个实现一个自适应尺寸的输出。（传统的池化层如最大池化、平均池化的输出大小是和输入大小挂钩的，但是我们最后做全连接层实现分类的时候需要指定全连接的输入，所以我们需要一种方法让神经网络在某层得到一个固定维度的输出，而且这种方法最好不是resize（resize会失真），由此SPP应运而生，其最早是何凯明提出，应用于RCNN模型）。当今的SPP在faster-rcnn上已经发展为今天的Multi-Scale-ROI-Align，而在Yolo上发展为SPPF。 - [yolov5n(SPPF)](https://github.com/WangRongsheng/BestYOLO/blob/main/models/yolov5n.yaml) - [yolov5n-SPP](https://github.com/WangRongsheng/BestYOLO/blob/main/models/SPP/yolov5n-SPP.yaml) - [yolov5n-SimSPPF](https://github.com/WangRongsheng/BestYOLO/blob/main/models/SPP/yolov5n-SimSPPF.yaml) - [yolov5n-ASPP](https://github.com/WangRongsheng/BestYOLO/blob/main/models/SPP/yolov5n-ASPP.yaml) - [yolov5n-RFB](https://github.com/WangRongsheng/BestYOLO/blob/main/models/SPP/yolov5n-RFB.yaml) - [yolov5n-SPPCSPC](https://github.com/WangRongsheng/BestYOLO/blob/main/models/SPP/yolov5n-SPPCSPC.yaml) - [yolov5n-SPPCSPC_group](https://github.com/WangRongsheng/BestYOLO/blob/main/models/SPP/yolov5n-SPPCSPC_group.yaml) - [yolov5n-SimCSPSPPF](https://github.com/WangRongsheng/BestYOLO/blob/main/models/SPP/yolov5n-SimCSPSPPF.yaml) |models|layers|parameters| |:-|:-|:-| |yolov5n(SPPF)|214|1766623| |yolov5n-SPP|217|1766623| |yolov5n-SimSPPF|216|1766623| |yolov5n-ASPP|214|3831775| |yolov5n-RFB|251|1932287| |yolov5n-SPPCSPC|232|3375071| |yolov5n-SPPCSPC_group|232|2047967| |yolov5n-SimCSPSPPF|229|3375071| - [yolov5n](https://github.com/WangRongsheng/BestYOLO/blob/main/models/yolov5n.yaml) - [yolov5n-FPN-AC](https://github.com/WangRongsheng/BestYOLO/blob/main/models/Attention/Self/yolov5n-FPN-AC.yaml) - [yolov5n-PAN-AC](https://github.com/WangRongsheng/BestYOLO/blob/main/models/Attention/Self/yolov5n-PAN-AC.yaml) - [yolov5n-FPN+PAN-AC](https://github.com/WangRongsheng/BestYOLO/blob/main/models/Attention/Self/yolov5n-FPN+PAN-AC.yaml) - [yolov5n-FPN-AS](https://github.com/WangRongsheng/BestYOLO/blob/main/models/Attention/Self/yolov5n-FPN-AS.yaml) - [yolov5n-PAN-AS](https://github.com/WangRongsheng/BestYOLO/blob/main/models/Attention/Self/yolov5n-PAN-AS.yaml) - [yolov5n-FPN+PAN-AS](https://github.com/WangRongsheng/BestYOLO/blob/main/models/Attention/Self/yolov5n-FPN+PAN-AS.yaml) |models|layers|parameters| |:-|:-|:-| |yolov5n|214|1766623| |yolov5n-FPN-AC|188|1858399| |yolov5n-PAN-AC|186|1642591| |yolov5n-FPN+PAN-AC|160|1734367| |yolov5n-FPN-AS|204|2106847| |yolov5n-PAN-AS|194|1891039| |yolov5n-FPN+PAN-AS|184|2231263| - [Optimal Transport Assignment](https://github.com/WangRongsheng/BestYOLO/blob/main/train.py#L476) - [辅助训练Optimal Transport Assignment](https://github.com/WangRongsheng/BestYOLO/blob/main/train_AuxOTA.py#L255) - [Soft-NMS](https://github.com/WangRongsheng/BestYOLO/blob/main/utils/general-softnms.py) > 训练不要使用`Soft-NMS`，耗时太久，请在`val`阶段开启，适用于小目标重叠数据。 - [Decoupled-head](https://blog.csdn.net/weixin_43694096/article/details/127427578) - [DCNv2](https://github.com/WangRongsheng/BestYOLO/blob/main/models/backbone/yolov5n-DCN.yaml) - [WBF](https://github.com/WangRongsheng/BestYOLO/blob/main/wbf.py) - [DCNv3](https://www.bilibili.com/video/BV1LY411z7iE) -