Mastering Computer Vision with TensorFlow 2.x

Encoder module

The key features of the encoder module are as follows:

• Atrous convolution to extract features.
• The output stride is the ratio of the input image's resolution to the final output resolution. A typical value for this is 16 or 8, which results in denser feature extraction.
• An Atrous convolution with a rate of 2 and 4 is used in the last two blocks.
• An ASPP module is used to apply convolution operations at multiple scales.