Abstract: The present disclosure describes techniques of improving video matting. The techniques comprise extracting features from each frame of a video by an encoder of a model, wherein the video comprises a plurality of frames; incorporating, by a decoder of the model, into any particular frame temporal information extracted from one or more frames previous to the particular frame, wherein the particular frame and the one or more previous frames are among the plurality of frames of the video, and the decoder is a recurrent decoder; and generating a representation of a foreground object included in the particular frame by the model, wherein the model is trained using segmentation dataset and matting dataset.

Abstract: Various embodiments are disclosed for simultaneous object localization and attribute classification using multitask deep neural networks. In an embodiment, a method comprises: obtaining, by a processing circuit, an image from an image capture device in an environment, the image including a target object in the environment; generating, by the processing circuit, predictions from the image for the target object using a multitask deep neural network, the multitask deep neural network including a network trunk and side branches, the network trunk configured for multi-scale feature extraction guided by supervision information provided by the side branches during training of the multitask deep neural network, the side branches configured as learning task-specific classifiers; and using, by the processing circuit, the predictions to localize the target object in the environment and to classify the target object and at least one attribute of the target object.

Abstract: Various embodiments are disclosed for detecting, tracking and counting objects of interest in video frames using fusion of image and depth modalities. In an embodiment, a method comprises: obtaining multiple frames of stereo image pairs from an image capturing device; rectifying each frame of the stereo image pairs; computing a stereo disparity for each frame of the stereo image pairs; determining a first set of object detections in each frame using the computed stereo disparity; determining a second set of object detections in each left or right frame of the stereo image pair using one or more machine learning models; fusing the first and second sets of object detections; and creating tracks based on the fused object detections.

Abstract: Various embodiments are disclosed for detecting, tracking and counting objects of interest in video. In an embodiment, a method of detecting and tracking objects of interest comprises: obtaining, by a computing device, multiple frames of images from an image capturing device; detecting, by the computing device, objects of interest in each frame; accumulating, by the computing device, multiple frames of object detections; creating, by the computing device, object tracks based on a batch of object detections over multiple frames; and associating, by the computing device, the object tracks over consecutive batches.

Abstract: Various embodiments are disclosed for detecting, tracking and counting objects of interest in video. In an embodiment, a method of detecting and tracking objects of interest comprises: obtaining, by a computing device, multiple frames of images from an image capturing device; detecting, by the computing device, objects of interest in each frame; accumulating, by the computing device, multiple frames of object detections; creating, by the computing device, object tracks based on a batch of object detections over multiple frames; and associating, by the computing device, the object tracks over consecutive batches.

Abstract: Various embodiments are disclosed for detecting, tracking and counting objects of interest in video. In an embodiment, a method of detecting and tracking objects of interest comprises: obtaining, by a computing device, multiple frames of images from an image capturing device; detecting, by the computing device, objects of interest in each frame; accumulating, by the computing device, multiple frames of object detections; creating, by the computing device, object tracks based on a batch of object detections over multiple frames; and associating, by the computing device, the object tracks over consecutive batches.

Abstract: Various embodiments are disclosed for simultaneous object localization and attribute classification using multitask deep neural networks. In an embodiment, a method comprises: obtaining, by a processing circuit, an image from an image capture device in an environment, the image including a target object in the environment; generating, by the processing circuit, predictions from the image for the target object using a multitask deep neural network, the multitask deep neural network including a network trunk and side branches, the network trunk configured for multi-scale feature extraction guided by supervision information provided by the side branches during training of the multitask deep neural network, the side branches configured as learning task-specific classifiers; and using, by the processing circuit, the predictions to localize the target object in the environment and to classify the target object and at least one attribute of the target object.

Abstract: Various embodiments are disclosed for detecting, tracking and counting objects of interest in video frames using fusion of image and depth modalities. In an embodiment, a method comprises: obtaining multiple frames of stereo image pairs from an image capturing device; rectifying each frame of the stereo image pairs; computing a stereo disparity for each frame of the stereo image pairs; determining a first set of object detections in each frame using the computed stereo disparity; determining a second set of object detections in each left or right frame of the stereo image pair using one or more machine learning models; fusing the first and second sets of object detections; and creating tracks based on the fused object detections.

Abstract: Various embodiments are disclosed for detecting, tracking and counting objects of interest in video. In an embodiment, a method of detecting and tracking objects of interest comprises: obtaining, by a computing device, multiple frames of images from an image capturing device; detecting, by the computing device, objects of interest in each frame; accumulating, by the computing device, multiple frames of object detections; creating, by the computing device, object tracks based on a batch of object detections over multiple frames; and associating, by the computing device, the object tracks over consecutive batches.

Abstract: A method for counting individuals in an image containing a dense, uniform or non-uniform crowd. The current invention leverages multiple sources of information to compute an estimate of the number of individuals present in a dense crowd visible in a single image. This approach relies on multiple sources, such as low confidence head detections, repetition of texture elements (using SIFT), and frequency-domain analysis to estimate counts, along with confidence associated with observing individuals in an image region. Additionally, a global consistency constraint can be employed on counts using Markov Random Field. This caters for disparity in counts in local neighborhoods and across scales. The methodology was tested on a new dataset of fifty (50) crowd images containing over 64,000 annotated humans, with the head counts ranging from 94 to 4,543. Efficient and accurate results were attained.

Abstract: A method for counting individuals in an image containing a dense, uniform or non-uniform crowd. The current invention leverages multiple sources of information to compute an estimate of the number of individuals present in a dense crowd visible in a single image. This approach relies on multiple sources, such as low confidence head detections, repetition of texture elements (using SIFT), and frequency-domain analysis to estimate counts, along with confidence associated with observing individuals in an image region. Additionally, a global consistency constraint can be employed on counts using Markov Random Field. This caters for disparity in counts in local neighborhoods and across scales. The methodology was tested on a new dataset of fifty (50) crowd images containing over 64,000 annotated humans, with the head counts ranging from 94 to 4,543. Efficient and accurate results were attained.

Abstract: Various embodiments are disclosed for detecting, tracking and counting objects of interest in video. In an embodiment, a method of detecting and tracking objects of interest comprises: obtaining, by a computing device, multiple frames of images from an image capturing device; detecting, by the computing device, objects of interest in each frame; accumulating, by the computing device, multiple frames of object detections; creating, by the computing device, object tracks based on a batch of object detections over multiple frames; and associating, by the computing device, the object tracks over consecutive batches.