Data and code for article "Nature reserve customized method of photo and video camera traps materials processing using two-stage neural network approach"

<strong>DESCRIPTION</strong> 📓 "data" folder directory contains the datasets for classification and detection. The detection dataset has <strong>YOLOv5 format</strong> and contains three classes <strong>[tigers, leopards, empty]</strong>. The class empty is about <strong>10%</strong> of the total data. The leopard and tiger classes contain <strong>3500</strong> images each. The entire amount of data for the detection task is <strong>7600</strong> images. The classification dataset contains two classes <strong>[tigers, leopards]</strong>. Images for classification are cropped images from the detection task using bounding boxes. Each class has <strong>3500</strong> images The "weights" folder contains pretrained models for classification and detection tasks. The detector weights were pre-trained on <strong>231k</strong> images from camera traps located throughout Russia. The classifier weights were pre-trained on <strong>416k</strong> images that were cropped with <strong>bounding boxes</strong> from photographs for the detection task. Some of the images for the classification task were taken from the <strong>Internet</strong>. The classifiers were trained for <strong>29 classes</strong>. You can also find folder <strong>tigers_vs_leopards</strong> in both the detection and classification directory, where there are weights that have been trained on a part of the camera trap images available at the link below. <em>Classification weights</em> EfficientNetv2-M <strong>ResNeSt-101e</strong> (🚀 RECOMMENDED) ResNet-101d ReXnet-100 SeResNet-152d <em>Detection weights</em> YOLOR-W6-1280 YOLOX-X-640 YOLOv5-X-640 YOLOv5-X-1280 YOLOv5-M6-1280 <strong>YOLOv5-L6-1280</strong> (🚀 RECOMMENDED) Read README.md file for more details**数据集说明** 📓 `data`文件夹内含分类与检测任务所需的数据集。检测数据集采用**YOLOv5格式**，涵盖3个类别：**[虎、豹、空背景]**。空背景类样本占检测任务总数据的约10%；豹类与虎类样本各含3500张图像，检测任务总数据量为7600张图像。 分类数据集涵盖2个类别：**[虎、豹]**。分类任务所用图像均为从检测任务样本中通过边界框（bounding boxes）裁剪得到的子图像，每个类别各含3500张图像。 `weights`文件夹内含分类与检测任务的预训练模型。检测模型权重基于全俄布设的红外相机陷阱（camera traps）采集的**23.1万**张图像预训练而来；分类模型权重则基于检测任务照片中经边界框裁剪得到的**41.6万**张图像预训练而成。部分分类任务图像来源于**互联网**，该分类模型原训练任务涵盖**29个类别**。 在检测与分类目录中均可找到名为**`tigers_vs_leopards`**的文件夹，内含基于下述链接提供的部分红外相机陷阱图像训练得到的模型权重。 *分类模型权重*：EfficientNetv2-M、ResNeSt-101e（🚀 推荐）、ResNet-101d、ReXnet-100、SeResNet-152d *检测模型权重*：YOLOR-W6-1280、YOLOX-X-640、YOLOv5-X-640、YOLOv5-X-1280、YOLOv5-M6-1280、**YOLOv5-L6-1280**（🚀 推荐） 详见`README.md`文件。