On October 17, 2018, Tencent AI Lab announced the official open source project “Tencent ML-images”, the address is github.com/Tencent/ten… . The project consists of mL-images, a multi-label image dataset, and ResNET-101, a deep residual network with the highest accuracy of its kind in the industry.
The main contents of the open source project include:
1. All image URLs of the ML-images dataset, with corresponding category annotations. Due to the copyright of the original image, this open source will not provide the original image directly. Users can download the image by themselves using the download code and URLs provided by us.
2. Detailed introduction of ML-images data set, including image source, image quantity, category quantity, semantic label system of category, labeling method, image labeling quantity and other statistics.
3. Complete code and model. The code we provide covers the whole process from image download, image preprocessing, mL-image-based pre-training, imagenet-based transfer learning, to image feature extraction based on the trained model. This project provides training examples based on small data sets to facilitate users to quickly experience our training process. The project also provides a resNET-101 model with very high accuracy (top-1 accuracy of 80.73% on the validation set of the single-label benchmark data set ImageNet). Users can choose the code or model of the project according to their own needs.
“Tencent ML – Images” as Tencent open source 58th project (https://github.com/Tencent), which is Tencent AI Lab accumulated in the field of computer vision based ability once released, It will provide researchers and engineers in the field of artificial intelligence with sufficient high-quality training data and easy-to-use, powerful deep learning models. At work, ML-images can provide strong support for visual tasks including Images and videos, and help improve the technical level of image classification, object detection, object tracking, semantic segmentation, and promote the common development of the artificial intelligence industry.