1. Feature extraction: Before puncturing, the model usually performs feature extraction according to the input image data, such as edge detection and texture analysis. These features can be used for subsequent tasks such as classification and segmentation. However, after the model is punctured, it will do more in-depth analysis and processing on specific areas or objects, such as extracting the characteristics of specific areas and carrying out more complicated segmentation.
2. Target detection: Before the puncture, the model will detect the targets in the whole image, such as the objects and faces in the image. However, after puncture, the model will detect targets or objects in specific areas, such as detecting objects and faces in specific areas.