Abstract: A method for de-identifying a privacy-related region within an image, including steps of: (a) inputting an input image into a segmentation network to apply a segmentation operation to the input image and generate at least part of (i) each of region probabilities of each of the pixels being estimated as the privacy-related region and (ii) each of region sizes assigned to each of the pixels of the input image; and (b) (i) calculating each of standard deviations for each of the pixels of the input image by using at least part of each of the region probabilities and each of the region sizes, thereby generating each of region standard deviations and (ii) applying a de-identifying operation to each of the pixels of the input image by using each of the region standard deviations, to thereby de-identify the privacy-related region within the input image.

Abstract: Method of training an object detector for predicting centers of mass of objects projected onto a ground is provided. The method includes steps of: acquiring training images from training data set; inputting each of training images into the object detector to thereby instruct the object detector to perform object detection for the training images and thus generate object detection results including (i) information on predicted bounding boxes, corresponding to one or more ROIs, acquired by predicting each of locations of the objects in the training images and (ii) information on predicted projection points acquired by projecting the centers of mass of the objects onto the ground; and training the object detector by using object detection losses generated by referring to the object detection results and information on ground truths corresponding to the training images.

Abstract: A method of training a video object detection model by using a training dataset is provided, including steps of: a learning device (a) after acquiring a training image (i) inputting the training image and first prior information, set as having probabilities of objects existing in locations in the training image, into the video object detection model, to thereby detect the objects and thus output first object detection information, and (ii) generating second prior information, which includes location information of the objects on the training image; (b) inputting the training image and the second prior information into the video object detection model, to detect the objects on the training and thus output second object detection information; and (c) generating a loss by referring to the second object detection information and a ground truth corresponding to the training image, and train the video object detection model to minimize the loss.

Abstract: A method for tracking one or more objects in a specific space is provided. The method includes steps of: (a) inputting original images of the specific space taken from camera to an obfuscation network and instructing the obfuscation network to obfuscate the original images to generate obfuscated images such that the obfuscated images are not identifiable as the original images by a human but the obfuscated images are identifiable as the original images by a learning network; (b) inputting the obfuscated images into the learning network, and instructing the learning network to detect obfuscated target objects, corresponding to target objects to be tracked, in the obfuscated images, to thereby output information on the obfuscated target objects; and (c) tracking the obfuscated target objects in the specific space by referring to the information on the obfuscated target objects.

Abstract: A learning method for training an obfuscation network capable of obfuscating original data for privacy, including steps of: (a) inputting training data into the obfuscation network to filter frequency information of the training data and thus generate obfuscated data; and (b) (i) inputting the obfuscated data into a learning network to generate characteristic information by performing learning operation on the obfuscated data, (ii) generating at least one task loss by referring to (ii-1) the characteristic information and its corresponding ground truth or (ii-2) a task-specific output, generated by using the characteristic information, and its corresponding ground truth, and (iii) training at least one of the learning network and the obfuscation network through a backpropagation of the task loss.

Abstract: A method for training a homography attention module (HAM) to perform multi-view object detection includes steps of: generating, from an i-th feature map corresponding to each of multiple training images representing multi-views of a target space, a 1-st to a d-th channel attention map for determining channel attention scores each channel included in the i-th feature map has for each of a 1-st to a d-th height plane of the target space, generating a 1-st to a d-th channel refined feature map by referring to channels with top k channel attention scores for each height, element-wisely multiplying them with corresponding spatial attention map generated therefrom to produce a 1-st to a d-th spatial refined feature map, and then homographically transforming them onto corresponding height plane and aggregating them to generate a BEV occupancy heatmap, which is used with its GT for training.

Abstract: A training image to be used in training a learning network is generated. The method of generating the training image includes steps of: (a) a labeling device, in response to acquiring an original image, (i) inputting the original image into an image recognition network to detect privacy-related regions from the original image, (ii) adding dummy regions, different from the detected privacy-related regions, onto the original image, and (iii) setting the privacy-related regions and the dummy regions as obfuscation-expected regions which represent regions to be obfuscated in the original image; (b) the labeling device generating an obfuscated image by obfuscating the obfuscation-expected regions; and (c) the labeling device labeling the obfuscated image to be corresponding to a task of the learning network to be trained, to thereby generate the training image to be used in training the learning network.

Abstract: A method for tracking one or more objects in a specific space is provided. The method includes steps of: (a) inputting original images of the specific space taken from camera to an obfuscation network and instructing the obfuscation network to obfuscate the original images to generate obfuscated images such that the obfuscated images are not identifiable as the original images by a human but the obfuscated images are identifiable as the original images by a learning network; (b) inputting the obfuscated images into the learning network, and instructing the learning network to detect obfuscated target objects, corresponding to target objects to be tracked, in the obfuscated images, to thereby output information on the obfuscated target objects; and (c) tracking the obfuscated target objects in the specific space by referring to the information on the obfuscated target objects.

Abstract: A method for training an obfuscation network is provided. The method includes steps of: a learning device (a) inputting training data into an obfuscation network to generate obfuscated data for training; (b) (i) inputting the obfuscated data for training into a discriminator to output a current obfuscation score for training and (ii) (ii-1) inputting first sub-data for training into a learning network to output first sub characteristic information for training and updating current updated learning parameters of the learning network to first sub updated learning parameters and (ii-2) while increasing an integer k from 2 to n, inputting k-th sub-data for training into the learning network to output k-th sub characteristic information for training and updating (k?1)-th sub updated learning parameters of the learning network to k-th sub updated learning parameters; and (c) updating previous updated obfuscation parameters of the obfuscation network to current updated obfuscation parameters.

Abstract: Methods for training a universal discriminator and an obfuscation network are provided.

Abstract: A method for training a user learning network for recognizing obfuscated data is provided. The method includes steps of: a learning device (a) (i) inputting obfuscated data, from a data provider, into a user learning network to generate first characteristic information and (ii) updating parameters of a user task layer and a first user batch normalizing layer such that an error, calculated using (1) the first characteristic information or a first task specific output and (2) a first ground truth of the obfuscated data, is minimized, and (b) (i) inputting original data, from a user, into the user learning network to generate second characteristic information and (ii) updating parameters of the user task layer and the second user batch normalizing layer such that an error, calculated using (1) the second characteristic information or a second task specific output and (2) a second ground truth of the original data, is minimized.

Abstract: A method for training an obfuscation network for obfuscating data is provided.

Abstract: A method of training an obfuscation network for obfuscating original data to protect personal information is provided. The method includes steps of a learning device, (a) inputting acquired training data into an obfuscation network to obfuscate the training data and inputting the obfuscated training data into an augmentation network to augment the obfuscated training data; (b) (i) inputting the augmented obfuscated training data into a learning network to generate first characteristic information and (ii) inputting the training data into the learning network to generate second characteristic information; and (c) training the obfuscation network such that (i) a first error, calculated by using the first and the second characteristic information, is minimized and (ii) a second error, calculated by using (ii-1) modified training data or modified obfuscated training data, and (ii-2) the obfuscated training data or the augmented obfuscated training data, is maximized.

Abstract: A method for producing a labeled image is provided. The method includes steps of: a labeling server (i) providing an image modifying interface to a user device to generate at least one anonymized image by anonymizing the original image except a specific labeling region among at least one labeling region, or generate at least one cropped image by cropping the labeling region, thus generating at least one transformed image by applying at least one transform function to the anonymized image or the cropped image, (ii) acquiring an obfuscated image by obfuscating the original image, (iii) acquiring at least one partial labeled image by allowing labelers to label the transformed image, and (iv) inversely applying the transform function received from the user device to the partial labeled image, thus generating at least one piece of adjusted partial labeling information and combining thereof with the obfuscated image to generate the labeled image.

Abstract: A method for training an obfuscation network capable of performing distinct concealing processes for distinct regions of an original image is provided. The method includes steps of: a learning device (a) inputting a training image into the obfuscation network to generate an obfuscated training image by performing a 1-st to an n-th concealing process respectively on a 1-st to an n-th training region of the training image; (b) inputting the obfuscated training image into a 1-st to an n-th discriminator to respectively generate a 1-st to an n-th obfuscated image score on determining whether the obfuscated training image is real or fake, and inputting the obfuscated training image into an image recognition network to apply learning operation on the obfuscated training image to generate feature information for training; and (c) training the obfuscation network such that an accumulated loss is maximized, and an accuracy loss is minimized.

Abstract: A method of a master learning device to train obfuscation networks and surrogate networks is provided. The method includes steps of: a master learning device (a) acquiring obfuscated data and ground truths from learning devices corresponding to owners or delegates of the original data and their ground truths; (b) (i) inputting the obfuscated data into a surrogate network, to apply learning operation thereto and generate characteristic information, (ii) calculating losses using the ground truths and the characteristic information or its task specific output, and (iii) training the surrogate network such that the losses or their average is minimized; and (c) transmitting the losses to the learning devices, to train the obfuscation networks such that the losses are minimized and that other losses calculated using the original data and the obfuscated data are maximized, and transmit network gradients of the trained obfuscation networks to the master learning device for its update.

Abstract: A method for training an obfuscation network and a surrogate network is provided. The method includes steps of: a 1st learning device (a) inputting original data of a 1st party, corresponding thereto, into the obfuscation network to generate obfuscated data wherein the 1st party owns the original data or is an entity to whom the original data is delegated; (b) transmitting the obfuscated data and the ground truth to a 2nd learning device corresponding to a 2nd party, and instructing the 2nd learning device to (i) input the obfuscated data into the surrogate network to generate characteristic information, (ii) calculate 1st losses using the ground truth and one of the characteristic information and task specific outputs, and (iii) train the surrogate network minimizing the 1st losses, and transmit the 1st losses to the 1st learning device; and (c) training the obfuscation network minimizing the 1st losses and maximizing 2nd losses.

Abstract: A method for preventing breach of original data for deep learning is provided. The method includes steps of: a data breach preventing device (a) adding noise onto the acquired original data to generate 1-st noisy data; and (b)(b1) while increasing an integer k from 1 to an integer larger than 0, (i) inputting k-th noisy data into a learning network, to apply learning operations to the k-th noisy data using learned parameters of the learning network, and to output k-th characteristic information, and (ii) launching an adversarial attack on the k-th noisy data via backpropagation using at least one of (ii-1) (k_1)-st losses calculated using the k-th characteristic information and a 1-st ground truth, and (ii-2) (k_2)-nd losses calculated using (1) a k-th task specific output and (2) a 2-nd ground truth, and generating (k+1)-th noisy data, and (b2) as a result, generating n-th noisy data as watermarked data.

Abstract: A method for concealing original data to protect personal information is provided. The method includes steps of: a data obfuscation device (a) if the original data is acquired, inputting the original data or its modified data into a learning network, and allowing the learning network to (i) apply a network operation to the original data or the modified data using learned parameters of the learning network and thus to (ii) output characteristic information on the original data or the modified data; and (b) updating the original data or the modified data via backpropagation using part of (i) 1-st losses calculated by referring to the characteristic information and its corresponding 1-st ground truth, and (ii) 2-nd losses calculated by referring to (ii-1) a task specific output generated by using the characteristic information and (ii-2) a 2-nd ground truth corresponding to the task specific output, to thereby generate obfuscated data.

Abstract: A method for learning a data embedding network is provided. The method includes steps of: a learning device acquiring and inputting original training data and mark training data into the data embedding network which integrates them and generates marked training data; inputting the marked training data into a learning network which applies a network operation to them and generates 1-st characteristic information, and inputting the original training data into the learning network which applies a network operation to them and generates 2-nd characteristic information; learning the data embedding network such that a data error is minimized, by referring to part of errors referring to the 1-st and the 2-nd characteristic information and errors referring to task specific outputs and their ground truths, and a marked data score is maximized, and learning a discriminator such that a original data score is maximized and the marked data score is minimized.