Most computer vision tasks use a lot of data, soData enhancementIt is a technique that is often used to improve the performance of computer vision systems. The main problem of computer vision at the moment is that there is no way to get sufficient data, so no matter what you useTransfer learningOr someone elsePre-trained model, Or starting from the source code, data enhancement is a very feasible method. The following are common data enhancement methods in computer vision:
1. Vertical mirror symmetry
2. Random cropping: cropping can get different pictures in the data set, but it is not a perfect data enhancement method. In practice, this method is still very practical.
3. Rotation, distortion transformation (in practice, it is rarely used because it is too complicated)
4. Color conversion: add different distortion values to the RGB channels of the original image, so that there are in the training set Distorted picture. This operation will make the network more robust to the color of the photo
Computer Vision Data Enhancement
Guess you like
Origin blog.csdn.net/qq_42308217/article/details/109642224
Ranking