Systems and methods for geolocation prediction

A computer-implemented method for training an information extraction model to determine data representing a real-world location associated with an object depicted in a sequence of images, the information extraction model comprising:

• an image-feature extraction model;
• a weakly supervised object classification model trained using image data with weak classification labels;
• a geolocation prediction model;
• the method comprising, at a computing system comprising one or more processors:
• obtaining data representing a sequence of images with noisy classification, at least one of the sequence of images depicting the object;
• outputting, by the image-feature extraction model in response to inputting the sequence of images, a sequence of image features extracted from the data representing the sequence of images with noisy classification;
• outputting, by the object classification model in response to inputting the sequence of image features, classification data including a classification associated with the object depicted in the sequence of images and one or more classification labels associated with the sequence of images, wherein the classification data is determined based at least in part on temporal attention data and spatial attention data associated with the sequence of image features, the one or more temporal attention data and the spatial attention data being determined by the object classification model;
• training the object classification model based at least in part on the classification data and the noisy classification associated with the sequence of images;
• outputting, by the geolocation prediction model in response to inputting the sequence of image features, the temporal attention data, and the spatial attention data, the real-world location associated with the object depicted in the sequence of images;
• and
• training the geolocation prediction model using at least the sequence of image features, the temporal attention data, and the spatial attention data.

A computer-implemented method for extracting information from imagery, the method comprising:

• obtaining, at a computing system comprising one or more processors, data representing one or more images, at least one of the one or more of images depicting an object;
• inputting, by the computing system, the one or more images into a machine-learned information extraction model that is trained to extract location information from the one or more images;
• and
• obtaining as an output of the information extraction model in response to inputting the one or more images, by the computing system, data representing a real-world location associated with the object depicted in the one or more images, wherein
• the information extraction model includes a weakly supervised object classification model trained using image data with weak classification labels, and
• inputting, by the computing system, the one or more images into the machine-learned information extraction model that is trained to extract the location information from the one or more images comprises:
• inputting, by the computing system, a sequence of image features extracted from the data representing the one or more images, into the object classification model, and
• obtaining, as an output of the object classification model in response to inputting the sequence of image features, a classification associated with the object depicted in the one or more images and classification data including one or more classification labels associated with the object.