International Business Machines Corporation

Guo Qiang Hu

Jian Ding

Peng Ji

Jing Chang Huang

Jun Chi Yan

Peng Gao

Wei Shan Dong

Fan Li

Sheng Nan Zhu

Xin Zhang

Zhi Hu Wang

Chang Li

Yuan Yuan Ding

Ning Duan

Chun Ma

Xu Liang Li

Zi Yu Zhu

Xin Hao Cheng

Yi Min Wang

JingChang Huang

Jian Wang

Qi Cheng Li

Jian Xu

Jinfeng Li

Hao Chen

Zhi Jun Wang

Li Jun Mei

Li Ni Zhang

Jun Zhu

Bo Yang

Shiwan Zhao

Fan Jing Meng

Wei Zhao

Min Gong

Dong Sheng Li

Xinlin Wang

Liang Wu

Lei He

Zhe Yan

Fang Quan Xie

Stephen Mingyu Chu

Xiao Bo Li

Ting Yuan

Bo Wen Wei

Chang Chun Liu

A plurality of different images of a same region of interest in an object are input into a set of neural networks, wherein each image of the region has been captured under a different value of a variable condition. A classification for each image is generated by the set of neural networks, wherein each classification includes a confidence score in a prediction of whether a feature is present in the region. The image classifications are ensembled to generate a final classification for the region. By applying a loss function, a loss is computed based on comparing the final classification to a ground truth of whether the feature is present in the region. The parameters of the set of neural networks are adjusted based on the computed loss.

Feature detection based on neural networks

According to embodiments of the present invention, a method, a device and a computer program product for image processing is provided. A computing device obtains an image of a first object, the image presenting a defect of the first object. A computing device obtains defect distribution information indicating respective frequencies of a plurality of predetermined categories of defects presented at corresponding locations in a plurality of training images, the plurality of training images presenting second objects and being used for training a defect classifier. A computing device determines a target category of the defect of the first object by applying the image and the defect distribution information to the defect classifier. A computing device generates one or more correction notifications.

Object defect correction

Embodiments of the present invention facilitate product defect detection. A computer-implemented method comprises: receiving, by a device operatively coupled to one or more processors, a template image of a normal product; generating, by the device, one or more geometric training parameters for transforming the template image; and transforming, by the device, the template image using the one or more geometric training parameters to generate a transformed image for training a data model, wherein the trained data model being used for aligning the template image and an image under inspection of a product.

Product defect detection

In an approach to AI empowered factory automation using an industrial IoT, responsive to receiving a new production task, production requirements are input into an AI engine. Equipment and software containers are selected based on the AI model. An orchestration service is created, where the orchestration service collaborates the equipment and the software containers.

Artificial Intelligence empowered industrial Internet of Things

A method and system for generating a three-dimensional (3D) virtual scene are disclosed. The method includes: identifying a two-dimensional (2D) object in a 2D picture and the position of the 2D object in the 2D picture; obtaining the three-dimensional model of the 3D object corresponding to the 2D object; calculating the corresponding position of the 3D object corresponding to the 2D object in the horizontal plane of the 3D scene according to the position of the 2D object in the picture; and simulating the falling of the model of the 3D object onto the 3D scene from a predetermined height above the 3D scene, wherein the position of the landing point the model of the 3D object in the horizontal plane is the corresponding position of the 3D object in the horizontal plane of the 3D scene.

Generating three-dimensional virtual scene

A computer-implemented method applies labels to unlabeled public data for use by a global model. One or more processors train one or more local machine learning models with local private data to create one or more trained models. Processor(s) generate a label for each of the local private data using the one or more trained models, where each label describes the local private data, and then apply the label to unlabeled public data to create labeled public data. One or more processors then input the labeled public data into a global model that uses the public data.

Training a machine learning model in a distributed privacy-preserving environment

In an approach for object detection with missing annotations under visual inspection, a processor receives an image. A processor classifies the image being a not-good image using a pre-trained classifier. A not-good image means one or more defect objects being in the image. A processor, in response to classifying the image being the not-good image, detects the one or more defect objects in the not-good image. A processor masks the one or more defect objects in the not-good image. A processor inputs the masked image to train a detector.

Object detection with missing annotations in visual inspection

The present disclosure relates to training a machine learning model to classify images. An example method generally includes receiving a training data set including images in a first category and images in a second category. A convolutional neural network (CNN) is trained using the training data set, and a feature map is generated from layers of the CNN based on features of images in the training data set. A first area in the feature map including images in the first category and a second area in the feature map where images in the first category overlap with images in the second category are identified. The first category is split into a first subcategory corresponding to the first area and a second subcategory corresponding to the second area. The CNN is retrained based on the images in the first subcategory, images in the second subcategory, and images in the second category.

Classifying images in overlapping groups of images using convolutional neural networks

A method, a device and a computer program product for image processing are proposed. In the method, a first training image and region information are obtained. The region information indicates a region of a defect in the first training image. A second training image with the defect at least partially removed is generated using an image generator based on the first training image and the region information. The image generator is trained to recover the first training image by replacing pixels included in the region indicated by the region information. The image generator is updated based on the second training image. In this way, the image including the defect can be accurately and efficiently recovered.

Image recovery

Aspects described herein include a computer-implemented method and associated system and computer program product. The method includes training a model using a plurality of defect images. Each defect image corresponds to a respective first feature combination of encoded textual features of a predefined set of textual features. The method further includes generating a first synthetic image using the model. The first synthetic image corresponds to a second feature combination of encoded textual features of the predefined set that is distinct from the first feature combinations.

Generating synthetic defect images for new feature combinations

Methods, systems and computer program products for providing improved visualization of inspection results are provided. Aspects include receiving an image from a connected camera by a computing device and processing the image by a graphics processing unit to divide the image into a plurality of image blocks. Responsive to inputting the processed image into a trained model, a plurality of image areas within the processed image is identified. Each image area is associated with an importance level and includes a unique set of image blocks. A sequence of image blocks is determined based on an importance level associated with each of the plurality of image blocks, wherein the sequence includes a list of image blocks in descending order by the importance level of each image block. The image blocks are stored in order of the sequence of image blocks by the server.

Visualization of inspection results

Embodiments of the present disclosure relate to object defect detection. In an embodiment, a computer-implemented method is disclosed. According to the method, for a test image of at least one part of a target object, a reference image is generated by repeating a periodic pattern detected in the test image, the target object consisting of elements. A differential image is determined by comparing the test image and the reference image. The differential image is superimposed on a predefined grid image to obtain a superimposed image. The grid image comprises grids corresponding to elements of a reference object associated with the target object. The number of defective elements is determined in the at least one part of the target object based on the superimposed image. In other embodiments, a system and a computer program product are disclosed.

Object defect detection

A system, method and computer program product for retrieving trajectory data from huge datasets. In the method, there is received, at a processor device, a user query including a request for displaying trajectory data at a user device. From the user query, a query type and a current map visualization scale setting (mapscale) for visualizing the trajectory data on the user device display is determined. Then, responsive to a user query type and the determined mapscale setting, a corresponding reference level is selected. Based on the selected reference level, there is accessed from a memory storage device a data set of compressed trajectory data. This compressed trajectory data set is communicated to the user device, for presentation on the user device display. In one aspect, the system and method is adaptive, enabling storage and retrieval of trajectory data according to various degrees of visualization or granularity.

High performance and efficient multi-scale trajectory retrieval

A method, a device and a computer program product for image processing are proposed. In the method, whether a first image indicates a defect associated with a target object is determined. In response to determining that the first image indicates the defect, a second image absent from the defect is obtained based on the first image. The defect is identified by comparing the first image with the second image. In this way, the defect associated with the target object in the image can be accurately and efficiently identified or segmented.

Image defect identification

Growing community of inventors

Shanghai, China

Guo Qiang Hu