Abstract: The present disclosure relates to methods, systems, and computer program products for implementing a deep neural network in a field-programmable gate array (FPGA). In response to receiving a network model describing a deep neural network, a plurality of layers associated with the deep neural network may be determined. With respect to a layer in the plurality of layers, a parallelism factor for processing operations associated with the layer simultaneously by processing elements in an FPGA may be determined based on a workload associated with the layer and a configuration of the FPGA.

Abstract: Methods and systems for training a machine learning model include training a machine learning model using training data. A status of the machine learning model's training is determined based on an accuracy curve of the machine learning model over the course of the training. Parameters of the training are adjusted based on the status. Training of the machine learning model is completed using the adjusted parameters.

Abstract: According to one embodiment, a method, computer system, and computer program product for generating anchors and selecting feature maps for a multi-scale object detection program based on analysis of the dataset is provided. The present invention may include generating a scale distribution of one or more scales of ground-truth objects, and, based on the scale distribution, dividing the effective scale range into a number of anchors greater than zero; furthermore, the invention may include generating a ratio distribution of ratios of the ground-truth objects; based on the ratio distribution, generating a ratio for at least one of the number of anchors; determining a template scale of one or more feature maps; and assigning the number of anchors to the feature maps based on the relative size of a scale of an anchor matching the relative size of a template scale of a feature map.

Abstract: A computer-implemented method for attention generation is provided. In this method, a plurality of image frames can be obtained from a video stream. An original attention for a first image frame of the plurality of image frames can be generated. Then, at least one interested area can be identified in the first image frame. A local attention for each of the at least one interested area can be generated. Moreover, a total attention for the first image frame can be generated based on the original attention of the first image frame and the local attention of each of the at least one interested area.

Abstract: According to one embodiment, a method, computer system, and computer program product for generating anchors and selecting feature maps for a multi-scale object detection program based on analysis of the dataset is provided. The present invention may include generating a scale distribution of one or more scales of ground-truth objects, and, based on the scale distribution, dividing the effective scale range into a number of anchors greater than zero; furthermore, the invention may include generating a ratio distribution of ratios of the ground-truth objects; based on the ratio distribution, generating a ratio for at least one of the number of anchors; determining a template scale of one or more feature maps; and assigning the number of anchors to the feature maps based on the relative size of a scale of an anchor matching the relative size of a template scale of a feature map.

Abstract: A method, computer system, and computer program product for model selection for training a new dataset is provided. The present invention may include choosing a model from a set of models to be evaluated for training the new dataset, selecting a sample input from a subset of the new dataset, calculating a model activation score for each of the sample inputs in the chosen model, calculating an accumulated model activation score for the chosen model, depending on the model activation score of each of the sample inputs in the chosen model, calculating an accumulated model activation score for each model from the set of models to be evaluated for training the new dataset, and selecting the model for training the new dataset with the highest accumulated model activation score.

Abstract: The present disclosure provides a computer-implemented method, computer system and computer program product for object detection. According to the computer-implemented method, a first object can be classified to be a first category based on detection data acquired in a detection area. Then, a confidence score for the first category can be calculated based on historical detection data of the detection area, wherein the confidence score presents possibility degree of at least one object of the first category existing in the detection area. Whether classification of the first object is abnormal can be determined based on the confidence score.

Abstract: A method comprises receiving a first received signal strength indicator (RSSI) of a first beacon in an array of beacons and receiving a second RSSI of a second beacon in an array of beacons, calculating a RSSI of the array (r) as a function of the first RSSI and the second RSSI, retrieving a calibrated RSSI value of the array (r?) from a memory, determining whether r>r?, and outputting a signal to a user device responsive to determining that r>r?.

Abstract: A computer-implemented method for attention generation is provided. In this method, a plurality of image frames can be obtained from a video stream. An original attention for a first image frame of the plurality of image frames can be generated. Then, at least one interested area can be identified in the first image frame. A local attention for each of the at least one interested area can be generated. Moreover, a total attention for the first image frame can be generated based on the original attention of the first image frame and the local attention of each of the at least one interested area.

Abstract: A computer implemented method of detecting excessive customer wait times is provided. The method includes taking a headcount in a digital image obtained by a digital video camera of a monitored area, counting the number of bodies in the digital image, and rectifying the number of heads with the number of bodies to obtain a total count of persons. The method further includes determining which persons are moving and subtracting the moving persons from the total count of persons to obtain a still count, and determining which persons are workers and subtracting the workers from the still count to identify customers in the monitored area and obtain a customer count. The method further includes identifying the number of queues present in the monitored area, assigning each customer to a queue, and determining the wait time for each of the identified customers in each of the identified queues.

Abstract: A computer-implemented method for data augmentation is provided according an embodiment of the present disclosure. In the method, a first feature vector for input data may be obtained based on a first model. The input data may be clustered to a plurality of clusters. For each of the clusters, a second feature vector may be obtained based on the first model. Then, a similarity between the first feature vector and the second feature vector may be estimated for each of the clusters. At least one cluster of the plurality of clusters for which the similarity is lower than a threshold may be determined. Moreover, data augmentation may be performed to the at least one cluster.

Abstract: The present disclosure provides a computer-implemented method, computer system and computer program product for object detection. According to the computer-implemented method, a first object can be classified to be a first category based on detection data acquired in a detection area. Then, a confidence score for the first category can be calculated based on historical detection data of the detection area, wherein the confidence score presents possibility degree of at least one object of the first category existing in the detection area. Whether classification of the first object is abnormal can be determined based on the confidence score.

Abstract: A computer implemented method of detecting excessive customer wait times is provided. The method includes taking a headcount in a digital image obtained by a digital video camera of a monitored area, counting the number of bodies in the digital image, and rectifying the number of heads with the number of bodies to obtain a total count of persons. The method further includes determining which persons are moving and subtracting the moving persons from the total count of persons to obtain a still count, and determining which persons are workers and subtracting the workers from the still count to identify customers in the monitored area and obtain a customer count. The method further includes identifying the number of queues present in the monitored area, assigning each customer to a queue, and determining the wait time for each of the identified customers in each of the identified queues.

Abstract: Embodiments of embodiments of the present invention relate to generation of a training model using virtual dataset and probe training models. A computer-implemented method comprises: receiving, by a device operatively coupled to one or more processors, a user dataset for training; testing, by the device, the user dataset with one or more probe training models; and in response to a result of the testing being similar to an existing result of running the one or more probe training models on an existing virtual dataset, grouping, by the device, the user dataset with the existing virtual dataset.

Abstract: Methods and systems for training a machine learning model include training a machine learning model using training data. A status of the machine learning model's training is determined based on an accuracy curve of the machine learning model over the course of the training. Parameters of the training are adjusted based on the status. Training of the machine learning model is completed using the adjusted parameters.

Abstract: A method, system and computer program product for accelerating a deep neural network (DNN) in a field-programmable gate array (FPGA) are disclosed. The method includes receiving a DNN net file and weights, converting the received DNN net file to one or more source files, generating an executable FPGA bit file using the one or more source files, and downloading the executable FPGA bit file from the DNN conversion platform to the FPGA. Converting of the received DNN net file and the weights to the one or more source files can further include analyzing the DNN net file to identify a plurality of neural layers, decomposing one or more neural layers of the plurality of neural layers to one or more operation blocks, instantiating the one or more source files, based on the one or more operation blocks.

Abstract: The present disclosure relates to methods, systems, and computer program products for implementing a deep neural network in a field-programmable gate array (FPGA). In response to receiving a network model describing a deep neural network, a plurality of layers associated with the deep neural network may be determined. With respect to a layer in the plurality of layers, a parallelism factor for processing operations associated with the layer simultaneously by processing elements in an FPGA may be determined based on a workload associated with the layer and a configuration of the FPGA.

Abstract: A method comprises receiving a first received signal strength indicator (RSSI) of a first beacon in an array of beacons and receiving a second RSSI of a second beacon in an array of beacons, calculating a RSSI of the array (r) as a function of the first RSSI and the second RSSI, retrieving a calibrated RSSI value of the array (r?) from a memory, determining whether r>r?, and outputting a signal to a user device responsive to determining that r>r?.

Abstract: A method comprises receiving a first received signal strength indicator (RSSI) of a first beacon in an array of beacons and receiving a second RSSI of a second beacon in an array of beacons, calculating a RSSI of the array (r) as a function of the first RSSI and the second RSSI, retrieving a calibrated RSSI value of the array (r?) from a memory, determining whether r>r?, and outputting a signal to a user device responsive to determining that r>r?.

Abstract: A method comprises receiving a first received signal strength indicator (RSSI) of a first beacon in an array of beacons and receiving a second RSSI of a second beacon in an array of beacons, calculating a RSSI of the array (r) as a function of the first RSSI and the second RSSI, retrieving a calibrated RSSI value of the array (r?) from a memory, determining whether r>r?, and outputting a signal to a user device responsive to determining that r>r?.