LAION-5B: An open large-scale dataset for training next generation image-text models.
Christoph SchuhmannRomain BeaumontRichard VencuCade GordonRoss WightmanMehdi ChertiTheo CoombesAarush KattaClayton MullisMitchell WortsmanPatrick SchramowskiSrivatsa KundurthyKatherine CrowsonLudwig SchmidtRobert KaczmarczykJenia JitsevPublished in: CoRR (2022)
Keyphrases
- million images
- image dataset
- input image
- image content
- single image
- object detectors
- low level
- outdoor images
- multiscale
- random fields
- image features
- image segmentation
- web images
- image classification
- image representation
- high resolution
- information retrieval
- web scale
- image data
- test images
- bayesian framework
- image collections
- image set
- street view
- classifier training
- image retrieval
- image analysis
- training dataset
- edge detection
- statistical model
- image regions
- segmentation method
- feature points
- text mining
- pixel values
- region of interest
- image matching
- training set
- object detection
- keywords