Abstract: An electronic device and a method for training or applying a neural network model are provided. The method includes the following steps. An input data is received. Convolution is performed on the input data to generate a high-frequency feature map and a low-frequency feature map. One of upsampling and downsampling is performed to match a first size of the high-frequency feature map and a second size of the low-frequency feature map. The high-frequency feature map and the low-frequency feature map are concatenated to generate a concatenated data. The concatenated data is inputted to an output layer of the neural network model.

Abstract: The image recognition method includes: obtaining an image data stream, wherein the image data frame includes a current frame; performing image recognition on an object in the current frame to generate a first box corresponding to the current frame; detecting movement of the object to generate a second box corresponding to the current frame; and determining the object as a tracking target according to the first box and the second box.

Abstract: The invention provides a method of generating an image recognition model and an electronic device using the method. The method includes the following. A source image is obtained; a first image is cut out of a first region of the source image to generate a cut source image; a preliminary image recognition model is pre-trained according to feature data and label data, in which the feature data is associated with the cut source image, and the label data is associated with the first image; and the pre-trained preliminary image recognition model is fine-tuned to generate the image recognition model. The method of generating the image recognition model and the electronic device provided by the invention may correctly restore an input image.

Abstract: An electronic device and a method for screening a sample are provided. The method includes the following steps. N samples corresponding to a first object are received, in which the N samples include a first sample. N similarity vectors respectively corresponding to the N samples are calculated, in which the N similarity vectors include a first similarity vector corresponding to the first sample. The first similarity vector includes multiple first similarities between the first sample and each of the N samples except the first sample. The first sample is determined to be a representative sample of the first object in response to an average value of the first similarities of the first similarity vector being the maximum value among average values of N similarities respectively corresponding to the N similarity vectors.

Abstract: An electronic device and a method for training or applying a neural network model are provided. The method includes the following steps. An input data is received. Convolution is performed on the input data to generate a high-frequency feature map and a low-frequency feature map. One of upsampling and downsampling is performed to match a first size of the high-frequency feature map and a second size of the low-frequency feature map. The high-frequency feature map and the low-frequency feature map are concatenated to generate a concatenated data. The concatenated data is inputted to an output layer of the neural network model.

Abstract: The invention relates to an unmanned aerial vehicle (UAV) and an image recognition method applied to a UAV. The image recognition method includes: obtaining an image data stream, wherein the image data frame includes a current frame; performing image recognition on an object in the current frame to generate a first box corresponding to the current frame; detecting movement of the object to generate a second box corresponding to the current frame; and determining the object as a tracking target according to the first box and the second box. A moving object can be accurately identified, detected, and tracked according to the UAV and the image recognition method provided in embodiments of the invention.

Abstract: A model training method and an electronic device are provided. The method includes: obtaining a first image; masking at least one region in the first image to obtain a masked image; inputting the masked image to a first model to obtain a first generated image; training the first model according to the first generated image and the first image; training a second model according to the first generated image and the first image; and when the first model is trained to a first condition and the second model is trained to a second condition, completing the training for the first model. By means of the model training method and the electronic device, the problem brought by a manually marked image can be resolved and the problem of causing mode collapse can be effectively avoided.