7 results about "Confidence value" patented technology

The invention discloses a scene text recognition method based on man-machine cooperation, and the method comprises the following steps: S1, carrying out preliminary processing of an existing scene text data set, and selecting a pre-training data set, a training set and a test set from the scene text data set; S2, training an SEE network by using the pre-training data set to obtain a pre-training model; S3, predicting the unmarked training set by adopting a pre-training model, and dividing the unmarked training set into a Hard sample and an Easy sample according to the degree of confidence of aprediction label generated by the model for the unmarked training set; performing manual annotation on the Hard sample, performing pseudo annotation on the Easy sample by using a model, and then performing fine tuning on the scene text recognition model by using the annotated sample; and S4, repeating the step S3 until the performance of the model meets the expected requirements. According to themethod, the marking cost of the scene text data set can be reduced, and the character recognition model performance can be improved.

The invention relates to the technical field of computers, in particular to a training data processing method and device, and the method comprises the steps of obtaining an original training data set, each training sample contained in the original training data set being respectively corresponding to at least two candidate marks; respectively determining the mark confidence coefficient of each candidate mark corresponding to each training sample, and respectively filtering each candidate mark corresponding to each training sample according to the mark confidence coefficient of each candidate mark; sampling each training sample corresponding to each filtered candidate mark so as to enable the number of training samples corresponding to different types of candidate marks to meet a preset number difference condition, and obtaining a target training data set; and using the target training data set for model training, so that the reliability and the balance of the training data are improved, and the accuracy of model training is further improved.

The invention discloses a user ID association method based on a big data technology. The method comprises the following steps: step A, reading configuration information; step B, according to the configuration information, pulling data from a data warehouse, and constructing vertexes and edges; step C, combining the repetitive items of the vertexes and the edges of the graph through SparkGraphx; step D, searching N-degree relations of vertexes of the graph in a directed manner and collecting the N-degree relations; step E, according to the N-degree relation node of each graph vertex collected in the previous step; step F, conducting confidence coefficient calculation on the ambiguous nodes collected in the last step according to a configured ambiguity rule, and completing affiliation judgment of ambiguous vertexes; and step G, according to the storage configuration, finally storing the output result in a specified database according to a specified format. The user ID association methodprovided by the invention has the advantages of more complete and more accurate mining data.

The invention discloses an image confrontation watermark generation method based on a particle swarm algorithm, and the method comprises the steps: combining an original image with a watermark through alpha mixing according to the position and transparency information of the watermark, and generating a watermark image; inputting the generated watermark image into a target neural network to obtain information of each classification confidence output by the target neural network; and on the basis of each classification confidence coefficient for the watermark image, obtaining an anti-watermark image through a particle swarm algorithm. According to the method, an adversarial watermark generation problem is modeled into an optimization problem, and the adversarial watermark is generated through a result of the optimization problem. And the optimization problem is efficiently solved by adopting a particle swarm algorithm. And a confidence coefficient difference value is used as an optimization target, so that omission of countermeasure watermark generation is avoided to a certain extent.

The invention provides an object three-dimensional shape reconstruction method and device, equipment and a storage medium, and relates to artificial intelligence image processing, and the method comprises the steps: carrying out the normalization of a two-dimensional image comprising a single object, and obtaining a normalized image; inputting the normalized image into an object shape reconstruction network, obtaining corresponding feature information through a feature extractor of the object shape reconstruction network, and inputting the feature information into a generator of the object shape reconstruction network to obtain a confidence three-dimensional tensor of a predicted voxel of the object; and carrying out binarization processing on each element in the confidence coefficient three-dimensional tensor based on a preset threshold value to obtain voxels representing the three-dimensional shape of the object. According to the method, the three-dimensional structure of the object can be accurately and efficiently recovered from the two-dimensional image, the efficiency of three-dimensional shape reconstruction is improved, and manpower and material resources are saved.