公开(公告)号：US20220398834A1

公开(公告)日：2022-12-15

申请号：US17820321

申请日：2022-08-17

Applicant: BEIJING BAIDU NETCOM SCIENCE TECHNOLOGY CO., LTD. ,  State Key Laboratory of Internet of Things for Smart City (University of Macau)

Inventor： Xingjian LI ,  Hang HUA ,  Chengzhong XU ,  Dejing DOU

IPC: G06V10/774

Abstract: A method for transfer learning includes: obtaining a pre-trained model, and generating a model to be transferred based on the pre-trained model, in which the model to be transferred includes N Transformer layers, and N is a positive integer; obtaining a mini-batch by performing random sampling on a target training set; and training the model to be transferred based on the mini-batch, in which a loss value for each Transformer layer is generated based on an empirical loss value and a noise stability loss value.

公开(公告)号：US20230072240A1

公开(公告)日：2023-03-09

申请号：US17988168

申请日：2022-11-16

Applicant: BEIJING BAIDU NETCOM SCIENCE TECHNOLOGY CO., LTD.

Inventor： Kafeng WANG ,  Chengzhong XU ,  Haoyi XIONG ,  Xingjian LI ,  Dejing DOU

IPC: G06K9/62

Abstract: A method for processing synthetic features is provided, and includes: the synthetic features to be evaluated and original features corresponding to the synthetic features are obtained. A feature extraction is performed on the synthetic features to be evaluated based on a number S of pre-trained samples, to obtain meta features with S samples. S is a positive integer. The meta features are input into the pre-trained meta feature evaluation model for a binary classification prediction, to obtain a probability of binary classification. Quality screening is performed on the synthetic features to be evaluated according to the probability of the binary classification, to obtain second synthetic features to be evaluated. The second synthetic features are classified in a good category. The second synthetic features and original features are input into a first classifier for evaluation. classified in a poor category.

公开(公告)号：US20220392199A1

公开(公告)日：2022-12-08

申请号：US17819777

申请日：2022-08-15

Applicant: BEIJING BAIDU NETCOM SCIENCE TECHNOLOGY CO., LTD. ,  State Key Laboratory of Internet of Things for Smart City (University of Macau)

Inventor： Kafeng WANG ,  Chengzhong XU ,  Haoyi XIONG ,  Xingjian LI ,  Dejing DOU

IPC: G06V10/774 ,  G06V10/764 ,  G06V10/82 ,  G06V10/778

Abstract: A method and an apparatus for training a classification model and data classification includes: obtaining a sample set and a pre-trained classification model, wherein the classification model includes at least two convolutional layers, each convolutional layer is connected to a classification layer through a fully connected layer; inputting the sample set into the classification model, and obtaining a prediction result output by each classification layer, wherein the prediction result includes a prediction probability of a class to which each sample belongs; calculating a probability threshold of each classification layer based on the prediction result output by each classification layer; setting a prediction stopping condition for the classification mode according to the probability threshold of each classification layer.

公开(公告)号：US20230244932A1

公开(公告)日：2023-08-03

申请号：US18076501

申请日：2022-12-07

Applicant: Beijing Baidu Netcom Science Technology Co., Ltd.

Inventor： Ji LIU ,  Qilong LI ,  Yu LI ,  Xingjian LI ,  Yifan SUN ,  Dejing DOU

IPC: G06N3/08 ,  G06V10/82

CPC classification number: G06N3/08 ,  G06V10/82

Abstract: Provided are an image occlusion method, a model training method, a device, and a storage medium, which relate to the technical field of artificial intelligence, in particular, to the field of computer vision technologies and deep learning, and may be applied to image recognition, model training and other scenarios. The specific implementation solution is as follows: generating a candidate occlusion region according to an occlusion parameter; according to the candidate occlusion region, occluding an image to be processed to obtain a candidate occlusion image; determining a target occlusion region from the candidate occlusion region according to visual security and data availability of the candidate occlusion image; and according to the target occlusion region, occluding the image to be processed to obtain a target occlusion image. In this manner, the image to be processed is desensitized while the accuracy of target recognition is ensured.

公开(公告)号：US20210319262A1

公开(公告)日：2021-10-14

申请号：US17355347

申请日：2021-06-23

Applicant: Beijing Baidu Netcom Science Technology Co., Ltd.

Inventor： Xingjian LI ,  Haoyi XIONG ,  Dejing DOU

IPC: G06K9/62 ,  G06K9/46 ,  G06N20/00 ,  G06N5/04

Abstract: The present application provides a model training, image processing method, device, storage medium, and program product relating to deep learning technology, which are able to screen auxiliary image data with image data for learning a target task, and further fuse the target image data and the auxiliary image data, so as to train a built and to-be-trained model with the fusion-processed fused image data. This implementation can increase the amount of data for training the model, and the data for training the model is determined is based on the target image data, which is suitable for learning the target task. Therefore, the solution provided by the present application can train an accurate target model even if the amount of target image data is not sufficient.