This application discloses a method and an apparatus for determining a posture of a target object in an image, a device, and a non-transitory storage medium. In the method, a training model parameter of a convolutional neural network of the target object is obtained from a server. A real-time image of the target object is obtained. At least one first image block from the real-time image is identified. The at least one first image block is a local image of the real-time image. According to the training model parameter, a label image block matching the at least one first image block is determined. The label image block is a local image of a standard image of the target object. Furthermore, the posture of the target object is determined, by processing circuitry of a terminal device, according to the at least one first image block and the determined label image block.