Research map / Label-efficient and robust learning / Recognition and long-tail
Autoencoding beyond pixels using a learned similarity metric
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use discriminator features instead of pixel distance | CIFAR-10 image similarity · CIFAR-10 | k=5 nearest neighbor classification error: 66.02% -> 33.73% | Helped |
| Add GAN training to the VAE with a learned feature metric | VAE · LFW test set | mean squared error: 27.59 ± 1.42 -> 22.39 ± 1.16 | Helped |
| Add GAN training to the VAE with a learned feature metric | VAE · LFW test set | cosine similarity: 0.9030 -> 0.9114 | Helped |
| Use discriminator features instead of pixel-space reconstruction error (ablation) | VAE/GAN · CelebA 64×64 face images | visual fidelity | Helped |
| Use a pretrained GAN discriminator as the feature metric (ablation) | VAE · CelebA 64×64 face images | visual fidelity | Mixed |
| Add the GAN objective to the decoder/generator (ablation) | VAE with discriminator-feature reconstruction · CelebA 64×64 face images | visual fidelity | Helped |
| Increase the relative contribution of discriminator-feature reconstruction (ablation) | VAE/GAN decoder · CelebA 64×64 face images | reconstruction fidelity | Hurt |
| Increase the relative contribution of the GAN objective (ablation) | VAE/GAN decoder · CelebA 64×64 face images | reconstruction detail | Hurt |
| Add the GAN objective to a VAE | VAE · LFW test set | mean squared error: 27.59 ± 1.42 -> 27.89 ± 3.07 | Hurt |
Closest papers by meaning
- Adversarial Latent Autoencoders (CVPR 2020 · similarity 0.89)
- Improving Visual Representation Learning Through Perceptual Understanding (CVPR 2023 · similarity 0.89)
- A Primal-Dual link between GANs and Autoencoders (NeurIPS 2019 · similarity 0.88)
- Diffusion Autoencoders: Toward a Meaningful and Decodable Representation (CVPR 2022 · similarity 0.87)
- Variational Lossy Autoencoder (ICLR 2017 · similarity 0.87)
- Rethinking the Paradigm of Content Constraints in Unpaired Image-to-Image Translation (AAAI 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.