Scaling Up Visual and Vision-Language Representation Learning With Noisy Text Supervision
By Chao Jia, Yinfei Yang, Ye Xia, Yi-Ting Chen, Zarana Parekh, Hieu Pham, Quoc V. Le, Yun-Hsuan Sung, Zhen Li, Tom Duerig. 2021
5,764 citations recorded by the source corpus.
Pre-trained representations are becoming crucial for many NLP and perception tasks. While representation learning in NLP has transitioned to training on raw text without human annotations, visual and vision-language representations still rely heavily on curated training datasets that are expensive or require expert knowledge. For vision applications, representations are mostly learned using datasets with explicit class labels such as ImageNet or OpenImages. For vision-language, popular datasets like Conceptual Captions, MSCOCO, or CLIP all involve a non-trivial data collection (and cleaning) process. This costly curation process limits the size of datasets and hence hinders the scaling of trained models. In this paper, we leverage a noisy dataset of over one billion image alt-text pairs, obtained without expensive filtering or post-processing steps in the Conceptual Captions dataset. A simple dual-encoder architecture learns to align visual and language representations of the image and text pairs using a contrastive loss. We show that the scale of our corpus can make up for its noise and leads to state-of-the-art representations even with such a simple learning scheme. Our visual representation achieves strong performance when transferred to classification tasks such as ImageNet and VTAB. The aligned visual and language representations enables zero-shot image classification and also set new state-of-the-art results on Flickr30K and MSCOCO image-text retrieval benchmarks, even when compared with more sophisticated cross-attention models. The representations also enable cross-modality search with complex text and text + image queries.
Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Multimodal Learning, Representation & Self-Supervised Learning
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
- Yoshua Bengio
- Percy Liang
- Luke S. Zettlemoyer
- Sergey Levine
- Caiming Xiong
In Computer Vision and Pattern Recognition
- Luc Van Gool
- Dacheng Tao
- Ming-Hsuan Yang
- Trevor Darrell
- Xiaogang Wang
- Jitendra Malik
- Andrew Zisserman
- Cordelia Schmid
In Multimodal Learning
In Representation & Self-Supervised Learning
- Andrew Zisserman
- Trevor Darrell
- Ishan Misra
- Andrea Vedaldi
- Cordelia Schmid
- Yoshua Bengio
- Antonio Torralba
- Dacheng Tao
Citations in the corpus
- Contrastive Learning of Medical Visual Representations from Paired Images and Text (2020, methodology): “…original release in 2020, ConVIRT has directly inspired subsequent studies such as the CLIP framework (Radford et al., 2021) and the ALIGN model (Jia et al., 2021), which showed that direct adaptations of ConVIRT-style pretraining at much larger scales lead to state-of-the-art general visual…”
- Do Image Classifiers Generalize Across Time? (2019, background): “Contrastive Language-Image Pre-training (CLIP) Recent advancements in large scale contrastive learning has leveraged supervision from text to achieve high zero shot performance on down stream tasks [23, 15].”
- Towards Language-Free Training for Text-to-Image Generation (2021, background): “This idea is recently scaled up in CLIP/ALIGN [15,35], which pave the way toward building a universal image-text representation space.”
- LAION-400M: Open Dataset of CLIP-Filtered 400 Million Image-Text Pairs (2021, background): “Multi-modal language-vision models demonstrated recently strong transfer capability to novel datasets in absense of per-sample labels [1, 2, 3].”
- SLIP: Self-supervision meets Language-Image Pre-training (2021, neutral): “Developed concurrently, ALIGN [31], uses a larger but noisier uncurated dataset and shows similar results.”
- Socratic Models: Composing Zero-Shot Multimodal Reasoning with Language (2022, background): “, alt-text captions (Jia et al. 2021)), or unifying model architectures (Hu and Singh 2021), Socratic Models1 (SMs) are a class of systems that embrace the zero-shot capabilities of foundation models by engineering guided discussions between the independent models to reach a shared consensus on a ta”
- Flamingo: a Visual Language Model for Few-Shot Learning (2022, methodology): “For our image and text pairs we first leverage the ALIGN [50] dataset, composed of 1.”
- Segment Anything (2023, methodology): “For example, CLIP [82] and ALIGN [55] use contrastive learning to train text and image encoders that align the two modalities.”
- 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).”
- BLIP: Bootstrapping Language-Image Pre-training for Unified Vision-Language Understanding and Generation (2022, methodology): “…methods (Chen et al., 2020; Li et al., 2020; 2021a; Wang et al., 2021; Radford et al., 2021) use image and alt-text pairs crawled from the web (Sharma et al., 2018; Changpinyo et al., 2021; Jia et al., 2021), Despite the use of simple rule-based filters, noise is still prevalent in the web texts.”
- MERLOT RESERVE: Neural Script Knowledge through Vision and Language and Sound (2022, methodology): “[62] Chao Jia, Yinfei Yang, Ye Xia, Yi-Ting Chen, Zarana Parekh, Hieu Pham, Quoc V Le, Yunhsuan Sung, Zhen Li, and Tom Duerig.”
- MAGMA - Multimodal Augmentation of Generative Models through Adapter-based Finetuning (2021, methodology): “On the other hand, contrastive models such as CLIP [37] and ALIGN [24] present a more robust approach to learning visual features by learning joint representations between image-text pairs.”
- Scaling Up Vision-Language Pretraining for Image Captioning (2021, methodology): “Besides CC12M, there also exist some other largescale image-text datasets, such as WIT [36], WenLan [16], LAION-400M [34], and the datasets used in CLIP [33] and ALIGN [17].”
- Florence: A New Foundation Model for Computer Vision (2021, methodology): “For all zero-shot tasks in this paper, we follow the setup in CLIP (Radford et al., 2021) and ALIGN (Jia et al., 2021) to remove near-duplicate test images from our training data.”