In semantic segmentation, when preprocessing data, both the image and the label are in the image format (height, width, channel). Your own data set is stored locally, you can use glob to read the address of each picture, use tensorflow to read the picture of each address, and then decode it.
- Get the address and convert it into a dataset object for easy processing,
- We can look at the shape and type of train, so we need to preprocess it later
- Preprocessing, decode and process the read pictures, and input them into the built network structure for training.
- After this data can be trained.
- Introduce glob, read all the subfolders and the file list under the subfolders that meet the requirements in the corresponding folder at a time, and glob.glob() returns the list....
