Synthetic Data from Diffusion Models Improves ImageNet Classification
By Shekoofeh Azizi, Simon Kornblith, Chitwan Saharia, Mohammad Norouzi, David J. Fleet. Trans. Mach. Learn. Res., 2023
451 citations recorded by the source corpus.
Deep generative models are becoming increasingly powerful, now generating diverse high fidelity photo-realistic samples given text prompts. Have they reached the point where models of natural images can be used for generative data augmentation, helping to improve challenging discriminative tasks? We show that large-scale text-to image diffusion models can be fine-tuned to produce class conditional models with SOTA FID (1.76 at 256x256 resolution) and Inception Score (239 at 256x256). The model also yields a new SOTA in Classification Accuracy Scores (64.96 for 256x256 generative samples, improving to 69.24 for 1024x1024 samples). Augmenting the ImageNet training set with samples from the resulting models yields significant improvements in ImageNet classification accuracy over strong ResNet and Vision Transformer baselines.
Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, General Social Sciences
View Paper: https://www.semanticscholar.org/paper/4538e353dd98f396c8facc29ebb72e9b1ba5f7c2
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Richard Zhang
- Phillip Isola
- Shobhita Sundaram
- Stephanie Fu
- Tali Dekel
- Netanel Y. Tamir
- Lucy Chai
- Ishan Misra
In Computer Vision and Pattern Recognition
- Richard Zhang
- Tali Dekel
- Phillip Isola
- Alexei A. Efros
- Luc Van Gool
- Ming-Hsuan Yang
- Andrea Vedaldi
- Jun-Yan Zhu
In General Social Sciences
Citations in the corpus
- Effective Data Augmentation With Diffusion Models (2023, unknown): “Our strategy is more appealing than fine-tuning the whole model as in Azizi et al. (2023) since it works from just one example per concept (Azizi et al. (2023) require millions of images), and doesn’t disturb the model’s ability to generate other concepts.”
- DreamSim: Learning New Dimensions of Human Visual Similarity using Synthetic Data (2023, methodology): “They have also been used to generate training data for a variety of downstream tasks [6, 9, 33, 75].”