The current CNN-based deep learning models for target detection are more mainstream based on supervised learning, such as the currently quite mature YOLO series, RCNN series, and centernet series target detection algorithm models. This supervised learning method is very dependent on the data set. The performance of model detection depends largely on the quantity and quality of the data set, that is, the more data, the more comprehensive, and the closer to the real application scenario, the better the trained model will perform in actual application. In the process of algorithm research, suitable data sets are also needed to verify the performance of algorithm design. Therefore, in the field of deep learning, data is of great significance. Many powerful companies are not only unique in algorithm design, but also unique in data retention.
Here we share a batch of data sets, let’s name it SATH data set, which is the name of the first letter of the target classification in English. It took me two years to find them bit by bit from the map, and then manually label them one by one. Thinking about keeping it for myself, it’s only of little use, why not make it public, if someone is willing to enjoy a cup of milk tea, it can be regarded as soothing the hair that fell out after staying up late. These data sets are all visible light imaging pictures, which can be used for artificial intelligence target detection model training, and can be downloaded by algorithm researchers for the purpose of algorithm research. The data set contains a total of four categories and the names of the categories, as shown in the figure below. There are more than these four types of data sets to be released in the future. The data set is being updated all the time, and the final amount of data may be 10,000 pictures of each type. A little bit of manual tagging is done, and every 1,000 copies will be published. The characteristics of the data set are all RGB color images, the size is 1024x1024, and the label format is xml format.
Dataset connection:
