Trending

Bookmarks

Keep Up With Latest Trending Papers. Computer Science, AI and more.Subscribe

Computer Vision

Computation and Language

RedCaps: A Large-scale Dataset of Images and Captions from Reddit

RedCaps: web-curated image-text data created by the people, for the people

We introduce a large-scale dataset of 12 million image-text pairs collected from reddit.

We show that captioning models trained on this dataset produce rich and varied captionspreferred by humans, and learn visual representations that transfer to many downstream tasks.

Authors

Karan Desai, Gaurav Kaul, Zubin Aysola, Justin Johnson

Related Topics

Visual representations

Captioning models

Querying search engines

Generic representations

Minimal filtering

Filtering pipelines

Read the Paper

Read Paper

Trending

Bookmarks

◐ Recommended Follows

More

◐ Latest Activity

More

RedCaps: A Large-scale Dataset of Images and Captions from Reddit - 42Papers

Read Related Papers

Multi-Label Iterated Learning

APANet: Adaptive Prototypes Alignment Network for Few-Shot Semantic Segmentation

Utilizing Resource-Rich Language Datasets for End-to-End Scene Text Recognition in Resource-Poor Languages

Tagger: Deep Unsupervised Perceptual Grouping

Decoupling Visual-Semantic Feature Learning for Robust Scene Text Recognition

ViCE: Self-Supervised Visual Concept Embeddings as Contextual and Pixel Appearance Invariant Semantic Representations

Hierarchical Modular Network for Video Captioning

Cerberus: An Attention-Based Architecture for Multi-Task Indoor Scene Understanding

Scaling Up Vision-Language Pre-training for Image Captioning

Mask-Image-Language trimodal encoder