advantages of PatchConvNet structure to abstract the image features

Improved accuracy: The PatchConvNet structure is designed to extract features from small image patches, which helps to capture more detailed information. This leads to improved accuracy in image classification tasks.

Robustness to variations: The PatchConvNet structure is able to handle variations in image appearance, such as changes in lighting, orientation, and scale. This is because it extracts features from small patches of the image, which are less affected by these variations.

Faster training: The PatchConvNet structure is faster to train compared to traditional convolutional neural networks because it requires fewer parameters. This is because it only processes small patches of the image instead of the entire image.

Scalability: The PatchConvNet structure can easily be scaled up to handle larger images by increasing the number of layers or the size of the patches.

Generalization: The PatchConvNet structure is able to generalize well to new images because it learns features that are relevant to the task rather than relying on hand-crafted features. This makes it useful for a wide range of computer vision applications