Reconstructed image
A reconstructed image is an output produced by decoding a compressed or latent representation to approximate an original input image.
Math and Architectures of Deep Learning
The output produced by an autoencoder that approximates the original input after decoding from a latent representation.
- How it is used
- Explains how autoencoders preserve essential image information through a compressed latent representation.
- How deep it goes
- Defines the decoder output, explains lossy encoding and reconstruction loss, and connects the concept to PyTorch implementations and VAE training.
- Examples
- Uses convolutional and transposed-convolutional networks, normalized pixel comparisons, MSE training code, and VAE reconstruction.
- What you can do after
- Understand how to interpret and train reconstructed-image outputs using latent vectors and reconstruction losses.
Read how Math and Architectures of Deep Learning explains reconstructed image →
Build a Text-to-Image Generator (from Scratch)
A reconstructed image is an image recovered by decoding a compressed or latent representation.
- How it is used
- Uses decoded outputs to verify whether compressed or quantized image representations preserve recognizable visual content.
- How deep it goes
- Traces encoding, latent compression, quantization, codebook lookup, decoding, and visual comparison, but does not derive training procedures.
- Examples
- Compares beaver, dolphin, cat, and fish images alongside their VAE or VQGAN reconstructions.
- What you can do after
- Apply pretrained VAE or VQGAN encoders and decoders to reconstruct images and inspect outputs with plotted comparisons.
Read how Build a Text-to-Image Generator (from Scratch) explains reconstructed image →
Related concepts
- VQGANproduced by · VQGAN is a generative model that combines vector-quantized discrete latent representations with adversarial and perceptual objectives to synthesize images.
- VQGAN decoderproduced by · A VQGAN decoder is the image-reconstruction component that maps discrete image tokens to an output image.
- decodeproduced by · Decoding is the process of transforming an encoded or internal representation into an interpretable or target-form output.
- reconstruction lossmeasured by · Reconstruction loss is a measure of the discrepancy between an input or observation and the output a model reconstructs from it or its representation.
- mean squared errormeasured by · Mean squared error is the arithmetic mean of the squared differences between corresponding predicted and target numerical values.
- decoder-only transformerproduced by · A decoder-only transformer is a Transformer architecture without a separate encoder that processes preceding context to predict or generate subsequent sequence elements.
These guides are generated from the text of Manning books by Manning's knowledge graph and link to the sections they draw on. The full books are available on liveBook and at manning.com.