Abstract: An apparatus may comprise a processor, and memory storing instructions, that when executed by the processor, may cause the apparatus to extract at least one source instance from at least one source image stored in a dataset, arrange the at least one extracted source instance on a target image stored in the dataset, wherein the target image may comprise at least one target instance associated with a ground-truth mask used to recognize the at least one target instance, perform, based on the arrangement, an augmentation process on the target image, adjust the ground-truth mask value based on the performance of the augmentation process, and output the target image including the adjusted ground-truth mask of the target instance.

Abstract: An object tracking apparatus based on meta learning receives inputs of a video frame and a target initialization sentence, specifies, when the input video frame is a first video frame of a video sequence, a target from the first video frame based on the target initialization sentence, detects, when the input video frame is not a first video frame, objects that may interact with the specified target from the input video frame, determines a target and a related object from the corresponding video frame by predicting an interaction between the specified target and the detected objects, determines the target and the related object until the last video frame of the video sequence, and predicts trajectories of the target and the related object.

Abstract: Object re-identification technology is used to track images, in which an object re-identification apparatus, an object re-identification method, and a model learning method can improve object re-identification performance through the use of a model trained based on spatially and temporally refined features. The object re-identification method includes extracting a first feature from an input video frame, and extracting a second feature by segmenting the first feature based on an expert module within multiple expert layers in a sequential relationship, where the expert module is a single expert module activated among multiple expert modules of a single expert layer.

Abstract: An instance segmentation model training method includes training the instance segmentation model firstly based on a first data set stored in a first database, and training the firstly-trained instance segmentation model secondly based on a second dataset stored in a second database, where the second dataset includes a large-scale open-source dataset and a segmentation target object-absent image acquired by capturing a working environment within an industrial site.

Abstract: An apparatus may comprise a processor, and memory storing instructions, that when executed by the processor, may cause the apparatus to extract at least one source instance from at least one source image stored in a dataset, arrange the at least one extracted source instance on a target image stored in the dataset, wherein the target image may comprise at least one target instance associated with a ground-truth mask used to recognize the at least one target instance, perform, based on the arrangement, an augmentation process on the target image, adjust the ground-truth mask value based on the performance of the augmentation process, and output the target image including the adjusted ground-truth mask of the target instance.

Abstract: A method of training a knowledge distillation based continual semantic segmentation model may include training the continual semantic segmentation model based on training data, predicting a probability for each class of an output of a current continual semantic segmentation model that has been generated by being trained and a probability for each class of an output of an old continual semantic segmentation model, expanding a class related to the old continual semantic segmentation model so that a spatial dimension of a class related to the old continual semantic segmentation model is the same as a spatial dimension of a class related to the current continual semantic segmentation model, calculating knowledge distillation loss based on the predicted probability for each class, and updating the current continual semantic segmentation model based on the knowledge distillation loss.