An apparatus includes a memory and a processor. The memory may be configured to store image data of an input image. The processor may be configured to detect one or more objects in the input image using a quantized multi-stage object detection network, where quantization of the quantized multi-stage object detection network includes (i) generating quantized image data by performing a first data range analysis on the image data of the input image, (ii) generating a feature map and proposal bounding boxes by applying a region proposal network (RPN) to the quantized image data, (iii) performing a region of interest pooling operation on the feature map and a plurality of ground truth boxes corresponding to the proposal bounding boxes generated by the RPN, (iv) generating quantized region of interest pooling results by performing a second data range analysis on results from the region of interest pooling operation, and (v) applying a region-based convolutional neural network (RCNN) to the quantized region of interest pooling results.

A method for recognizing an object in a video stream may include receiving a video stream from a video source, the video stream comprising a plurality of video frames. The method may also include selecting at least one video frame from the video frames according to a frame selection rate. The method may also include partitioning the selected video frame into a first plurality of image blocks. The method may also include recognizing, out of the first plurality of image blocks, a second plurality of image blocks which comprise an image of an object, the recognition being based on an image recognition parameter determined by a machine-learning algorithm. The method may also include determining that at least one of the second plurality of image blocks corresponds to the object based on a likelihood metric, the likelihood metric being determined by the processor based on at least the frame selection rate. The method may further include displaying, on a display, information identifying the object. A system and non-transitory computer-readable medium may also be provided.

A system and method for visually comparing one or more attributes of two fashion products on an e-commerce platform is presented. The system includes an image database, a selection module and a comparison module. The image database includes a plurality of two-dimensional (2D) images captured at the same scale for a plurality of fashion products. The selection module is configured to allow a user to select a target fashion product and a reference fashion product. The comparison module is configured to access a 2D image of the target fashion product and a 2D image of the reference fashion product from the image database; and allow the user to visually compare the target fashion product with the reference fashion product with respect to the one or more attributes by positioning the 2D image of the target fashion product on top or bottom of the 2D image of the reference fashion product.

A method and system for matching image features is applied to mine machine vision. The method includes de-noising an image, performing super-pixel segmentation to obtain a plurality of image blocks, calculating the information entropy of each image block to obtain an image block with information entropy greater than a first preset threshold, extracting feature points of the image block, to obtain the feature point set of the image, using a wavelet method to describe the feature points in the feature point set to obtain a feature point descriptor set, and matching feature points in the feature point set with the feature points of a target image.

A method for adjusting camera intrinsic parameters of a multi-camera visual tracking device is described. In one aspect, a method for calibrating the multi-camera visual tracking system includes disabling a first camera of the multi-camera visual tracking system while a second camera of the multi-camera visual tracking system is enabled, detecting a first set of features in a first image generated by the first camera after detecting that the temperature of the first camera is within the threshold of the factory calibration temperature of the first camera, and accessing and correcting intrinsic parameters of the second camera based on the projection of the first set of features in the second image and a second set of features in the second image.

Among other things, one or more systems and/or techniques for segmenting a representation of a sheet object from an image are provided herein. To identify elements of an image (e.g., pixels and/or voxels) representative of sheet objects, a constant false alarm rate (CFAR) score and a topological score are computed for respective elements being analyzed. The CFAR score indicates a relationship between an element and a neighborhood of elements when viewed as a collective unit. The topological score indicates a relationship between the element and a neighborhood of elements when viewed neighbor-by-neighbor. When the CFAR score is within a specified range of CFAR scores and the topological score is within a specified range of topological scores, the element is labeled as being associated with a sheet object. A connected component labeling (CCL) approach may be used to group elements labeled as being associated with a sheet object.

A method for training a multi-target image-text matching model and an image-text retrieval method are provided. The method for training the multi-target image-text matching model includes: obtaining a plurality of training samples that include sample pairs each including a sample image and a sample text, the sample image including a plurality of targets; obtaining, for each of the plurality of training samples, a heat map corresponding to the sample text in the training sample, the heat map representing a region of the target in the sample image that corresponds to the sample text; and training an image-text matching model based on a plurality of the sample texts and corresponding heat maps to obtain the multi-target image-text matching model.

Machine training and application of machine-trained classifier are used for image-based tumor phenotyping in a medical system. To create a training database with known phenotype information, synthetic medical images are created. A computational tumor model creates various examples of tumors in tissue. Using the computational tumor model allows one to create examples not available from actual patients, increasing the number and variance of examples used for machine-learning to predict tumor phenotype. A model of an imaging system generates synthetic images from the examples. The machine-trained classifier is applied to images from actual patients to predict tumor phenotype for that patient based on the knowledge learned from the synthetic images.

Systems and methods for operating an agricultural system are provided herein. The method can include presenting an image of a field on a display of an electronic device. The method can also include detecting a geographic position of the electronic device and an imager direction of the imager. In addition, the method can include determining a tilt orientation of the electronic device. The method can further include determining a field of view based on the geographic position, the tilt orientation, and the imager direction. The method can also include determining one or more features of an object within the image of the field and comparing the one or more features to stored feature data. Lastly, the method can include detecting a change in the one or more features of the object.

A method for mitigating motion blur in a visual-inertial tracking system is described. In one aspect, the method includes accessing a first image generated by an optical sensor of the visual tracking system, accessing a second image generated by the optical sensor of the visual tracking system, the second image following the first image, determining a first motion blur level of the first image, determining a second motion blur level of the second image, identifying a scale change between the first image and the second image, determining a first optimal scale level for the first image based on the first motion blur level and the scale change, and determining a second optimal scale level for the second image based on the second motion blur level and the scale change.