Picture classification model cold start method, system and device and medium
A technology of image classification and cold start, applied in biological neural network models, character and pattern recognition, instruments, etc., can solve problems such as difficulty in obtaining a large number of image classification models, achieve classification accuracy, realize data accumulation, and overcome training data hard-to-get effects
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0043] Combine below figure 1 The main flow chart of an embodiment of the method for cold-starting the image classification model according to the technical solution of the present invention is shown to illustrate the implementation process of the cold-start of the image classification model in the present invention. In this embodiment, the cold start of the CNN model is mainly helped by pre-SIFT feature matching.
[0044] Step S110, classify the pictures in the picture set based on SIFT feature matching.
[0045] For example, for a set of scene pictures in a target video, picture classification is performed based on SIFT feature matching.
[0046] Among them, several pictures can be selected from the target video in a predetermined time interval to form a scene picture set, and then the pictures in the scene picture set are classified through the SIFT feature matching algorithm to obtain the classified pictures. Wherein, the predetermined time interval is every second, and ...
example 1
[0066] Example 1: Taking a single classification as an example, for the first classification, the image is classified through SIFT feature matching, that is, the SIFT algorithm is used 100%; the classified image is used as training data to train the initial image classification model such as CNN0, and CNN1 is obtained. New model; for the second time, the SIFT algorithm and CNN0 can be used to continue to classify other unclassified pictures. The proportion of using the SIFT algorithm is 67%, while the proportion of using CNN0 is 33%; the classified pictures are used as training data for CNN1 That is, the new model (the model whose parameters have been updated relative to the last previous model, or the updated model) is trained to obtain CNN2, which is the new model; the third time still uses the SIFT algorithm and CNN1 to continue unclassified other pictures Classification, the proportion of using SIFT algorithm is 33%, while the proportion of using CNN1 is 33%; the classified...
example 2
[0067] Example 2: Taking a single classification as an example, it is also possible to classify pictures through SIFT feature matching during the first classification, that is, 100% use the SIFT algorithm; use the classified pictures as training data to train the initial picture classification model such as CNN0 , get CNN1 which is the new model; the second time you can use SIFT algorithm and CNN1 to continue to classify other unclassified pictures, the proportion of using SIFT algorithm is 67%, while the proportion of using CNN0 is 33%; the classified pictures are used as The training data trains CNN1, which is a new model relative to the initial model CNN0 (the model whose parameters have been updated relative to the last previous model, or the updated model), to obtain CNN2, which is the new model; the third time is still used at the same time The SIFT algorithm and CNN2 continue to classify other unclassified pictures. The proportion of using SIFT algorithm is 33%, while th...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More 


