We have proposed a large-scale & high-quality professional dataset -- FashionAI dataset, to facilitate the resolution to some fundamental fashion problems which AI techniques are applied to. This dataset consists of two tasks, apparel keypoints detection and apparel attributes recognition. With the endeavor of experts from both AI and fashion fields, we have made a clear definition of the problems in these tasks, and accordingly, created a knowledge system that is both professional and machine-learnable. On the basis of the knowledge structure, we collected image data, employed well-trained annotators to annotate them, and finally verified the annotation on a professional platform, thus constructing such an apparel dataset for practical scenarios. We expect the dataset to become an important data source and verification standard, to facilitates AI researchers' technological exploration in fashion field and promotes fashion-related technical innovation and fast product implementation.