A VQ-VAE decoder reconstructs an image from quantized latent codes: it uses each integer code to retrieve its codebook embedding, then maps the resulting vectors back into image space. section 11.3.2
The decoding path proceeds in three stages:
- Receive a sequence of discrete integer codes produced by quantization. section 11.3.2
- Retrieve the corresponding embeddings from the codebook. section 11.3.2
- Reconstruct the image from those retrieved latent vectors. section 11.3.2
Decoding a cat image
A quantized cat image may become 576 codes such as (873, 862, …, 148). Feeding the sequence to the decoder retrieves the indexed embeddings and reconstructs the cat image. section 11.3.2
Figure 11.5 The architecture of a VQ-VAE and its three components—encoder, codebook, and decoder—in the training process
Use the architecture figure to trace how the encoder, codebook, and decoder participate in VQ-VAE processing. section 11.3.2
