Removing Word-Level Spurious Alignment between Images and Pseudo-Captions in Unsupervised Image Captioning.
Ukyo HondaYoshitaka UshikuAtsushi HashimotoTaro WatanabeYuji MatsumotoPublished in: CoRR (2021)
Keyphrases
- input image
- image data
- image features
- image collections
- word level
- image retrieval
- image classification
- image regions
- image analysis
- preprocessing steps
- edge detection
- visual features
- image content
- web images
- latitude longitude
- language independent
- image segmentation
- image annotation
- character recognition
- document images
- machine learning
- machine translation
- semi supervised
- low level