A curated list of resources including papers, datasets, and relevant links pertaining to few-shot image generation. Since few-shot image generation is a very broad concept, there are various experimental settings and research lines in the realm of few-shot image generation.
The generative model is trained on base categories and applied to novel categories with (optimization-based) or without finetuning (fusion-based and transformation-based).
Optimization-based methods:
- Louis Clouâtre, Marc Demers: "FIGR: Few-shot Image Generation with Reptile." CoRR abs/1901.02199 (2019) [pdf] [code]
- Weixin Liang, Zixuan Liu, Can Liu: "DAWSON: A Domain Adaptive Few Shot Generation Framework." CoRR abs/2001.00576 (2020) [pdf] [code]
Fusion-based methods:
- Davis Wertheimer, Omid Poursaeed, Bharath Hariharan: "Augmentation-interpolative Autoencoders for Unsupervised Few-shot Image Generation." arXiv (2020). [pdf]
- Yan Hong, Li Niu, Jianfu Zhang, Weijie Zhao, Chen Fu, Liqing Zhang: "F2GAN: Fusing-and-Filling GAN for Few-shot Image Generation." ACM MM (2020) [pdf] [code]
- Yan Hong, Li Niu, Jianfu Zhang, Liqing Zhang: "MatchingGAN: Matching-based Few-shot Image Generation." ICME (2020) [pdf] [code]
- Sergey Bartunov, Dmitry P. Vetrov: "Few-shot Generative Modelling with Generative Matching Networks." AISTATS (2018) [pdf] [code]
Transformation-based methods:
- Antreas Antoniou, Amos J. Storkey, Harrison Edwards: "Data Augmentation Generative Adversarial Networks." stat (2018) [pdf] [code]
- Yan Hong, Li Niu, Jianfu Zhang, Jing Liang, Liqing Zhang: "DeltaGAN: Towards Diverse Few-shot Image Generation with Sample-Specific Delta." CoRR abs/2009.08753 (2020) [pdf]
Datasets:
- Omniglot: 1623 handwritten characters from 50 different alphabets. Each of the 1623 characters was drawn online via Amazon's Mechanical Turk by 20 different people [link]
- EMNIST: 47 balanced classes [link]
- FIGR: 17,375 classes of 1,548,256 images representing pictograms, ideograms, icons, emoticons or object or conception depictions [link]
- VGG-Faces: 2395 categories [link]
- Flowers: 8189 images from 102 flower classes [link]
- Animal Faces: 117574 images from 149 animal classes [link]
The generative model is trained on a large dataset (base domain/category) and finetuned on a small dataset (novel domain/category).
- Atsuhiro Noguchi, Tatsuya Harada: "Image generation from small datasets via batch statistics adaptation." ICCV (2019). [pdf] [code]
- Yaxing Wang, Abel Gonzalez-Garcia, David Berga, Luis Herranz, Fahad Shahbaz Khan, Joost van de Weijer: "MineGAN: effective knowledge transfer from GANs to target domains with few images." CVPR (2020). [pdf] [code]
- Yijun Li, Richard Zhang, Jingwan Lu, Eli Shechtman: "Few-shot Image Generation with Elastic Weight Consolidation." NeurIPS (2020). [pdf]
- Esther Robb, Wen-Sheng Chu, Abhishek Kumar, Jia-Bin Huang: "Few-Shot Adaptation of Generative Adversarial Networks." arXiv (2020). [pdf] [code]
- Miaoyun Zhao, Yulai Cong, Lawrence Carin: "On Leveraging Pretrained GANs for Generation with Limited Data." ICML (2020). [pdf] [code]
- Utkarsh Ojha, Yijun Li, Jingwan Lu, Alexei A. Efros, Yong Jae Lee, Eli Shechtman, Richard Zhang: "Few-shot Image Generation via Cross-domain Correspondence." CVPR (2021). [pdf] [code]
The generative model is directly trained on a small dataset.
- Shengyu Zhao, Zhijian Liu, Ji Lin, Jun-Yan Zhu, Song Han: "Differentiable Augmentation for Data-Efficient GAN Training." NeurIPS (2020). [pdf] [code]
- Bingchen Liu, Yizhe Zhu, Kunpeng Song, Ahmed Elgammal: "Towards Faster and Stabilized GAN Training for High-fidelity Few-shot Image Synthesis." ICLR (2021). [pdf] [code]
In the extreme case, the generative model is directly trained on a single image. However, the learnt model generally only manipulates the repeated patterns in this image.