3つの要点 ✔️ BigGANを超えた最強の非GANジェネレータの登場 ✔️ 階層型の潜在マップを使用し、潜在表現を異なるスケール毎に学習 ✔️ 階層内の各レベルを個別にピクセルに依存させることで、デコーダでの再構成エラーを減らす Generating Diverse High-Fidelity Images with VQ-VAE-2 written by Ali Razavi, Aaron van den Oord, Oriol Vinyals (Submitted on 2 Jun 2019) Comments: Published by arXiv Subjects: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (stat.ML) BigGAN
![BigGANを超えた!DeepMind、史上最強の非GANジェネレータVQ-VAE-2を提案](https://cdn-ak-scissors.b.st-hatena.com/image/square/c4d65cc58fa7bf0df0b61a9cbe3ee9e3a36bb4b8/height=288;version=1;width=512/https%3A%2F%2Faisholar.s3.ap-northeast-1.amazonaws.com%2Fmedia%2FJune2020%2F%25E7%2594%25BB%25E5%2583%258F%25E7%2594%259F%25E6%2588%2590%25E3%2583%25A2%25E3%2583%2587%25E3%2583%25AB%25E3%2581%25AE_%25E8%25A6%2587%25E6%25A8%25A9%25E4%25BA%2589%25E3%2581%2584_1_.png)