Scaling Vision Transformers
By Xiaohua Zhai, Alexander Kolesnikov, Neil Houlsby, Lucas Beyer. 2021
1,462 citations recorded by the source corpus.
Attention-based neural networks such as the Vision Transformer (ViT) have recently attained state-of-the-art results on many computer vision benchmarks. Scale is a primary ingredient in attaining excellent results, therefore, understanding a model's scaling properties is a key to designing future generations effectively. While the laws for scaling Transformer language models have been studied, it is unknown how Vision Transformers scale. To address this, we scale ViT models and data, both up and down, and characterize the relationships between error rate, data, and compute. Along the way, we refine the architecture and training of ViT, reducing memory consumption and increasing accuracy of the resulting models. As a result, we successfully train a ViT model with two billion parameters, which attains a new state-of-the-art on ImageNet of 90.45% top-1 accuracy. The model also performs well for few-shot transfer, for example, reaching 84.86% top-1 accuracy on ImageNet with only 10 examples per class.
Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Scaling & Learning Theory
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
- Jianfeng Gao
- Mohit Bansal
- Quoc V. Le
- Sergey Levine
- Yoshua Bengio
- Dacheng Tao
- Percy Liang
- Trevor Darrell
In Computer Vision and Pattern Recognition
- Luc Van Gool
- Dacheng Tao
- Ming-Hsuan Yang
- Trevor Darrell
- Xiaogang Wang
- Jitendra Malik
- Andrew Zisserman
- Alan L. Yuille
In Scaling & Learning Theory
- Neil Houlsby
- Soham De
- Xiaohua Zhai
- Basil Mustafa
- Mostafa Dehghani
- Yuandong Tian
- Quoc V. Le
- Samuel L. Smith
Citations in the corpus
- LAION-400M: Open Dataset of CLIP-Filtered 400 Million Image-Text Pairs (2021, background): “When increasing model and compute budget scale in addition, scaling laws suggest further increase in generalization and transfer performance if not bottlenecked by the data scale [5, 6, 7, 8].”
- Model soups: averaging weights of multiple fine-tuned models improves accuracy without increasing inference time (2022, methodology): “For our main results we fine-tune CLIP (Radford et al., 2021) and ALIGN (Jia et al., 2021), which are pre-trained with a contrastive loss on image-text pairs, and a ViT-G model pre-trained on JFT (Zhai et al., 2021).”
- Multiview Transformers for Video Recognition (2022, methodology): “In the second part of the tables, we compare to methods that are pretrained on web-scale datasets such as Instagram 65M [25], JFT-300M [62], JFT-3B [81], WTS [61], Florence [80] or HowTo100M [47].”
- Scaling Up Vision-Language Pretraining for Image Captioning (2021, methodology): “Recent work has investigated the Pareto frontier of training transformer models, often referred to as the neural scaling law, in the domains of natural language processing [2, 18, 41] and computer vision [12,47], via unsupervised or weakly-supervised learning methods.”
- Florence: A New Foundation Model for Computer Vision (2021, neutral): “, 2021), ViT-G/14 (Zhai et al., 2021), CoAtNet-7 (Dai et al.”
- Combined Scaling for Zero-shot Transfer Learning (2021, methodology): “Three well-known techniques to relieve memory burden are gradient accumulation (GradAccum) (Ott et al., 2018; Zhai et al., 2021), re-materialization (or gradient checkpointing) (Griewank and Walther, 2000; Chen et al., 2016) and model parallelism (Shazeer et al., 2018; Huang et al., 2019; Lepikhin…”
- LiT: Zero-Shot Transfer with Locked-image text Tuning (2021, methodology): “The pre-trained ViT-g/14 model [68], achieves 70.”
- Visual ChatGPT: Talking, Drawing and Editing with Visual Foundation Models (2023, methodology): “To better extract visual features, frozen pre-trained image encoders are adopted in the early works [9, 25, 54], and recent LiT [52] apply the CLIP pre-training [30] with frozen ViT model [51].”
- LAION-5B: An open large-scale dataset for training next generation image-text models (2022, methodology): “In the following decade, computer vision benefited from growth in model, data, and compute scale, in addition to advances in both convolutional and transformer architectures [15, 33, 81, 92].”
- Exploring Plain Vision Transformer Backbones for Object Detection (2022, background): “, block designs [49,50], self-supervised learning [1,22], and scaling [53].”
- Instance Segmentation for Autonomous Log Grasping in Forestry Operations (2022, methodology): “A linear regression on the semi-log plot, as depicted by the purple line in Figure 5a, shows that we are in the presence of a power law, as seen also by others using Vision Transformers [33].”