Unsafe Diffusion: On the Generation of Unsafe Images and Hateful Memes From Text-To-Image Models.
Yiting QuXinyue ShenXinlei HeMichael BackesSavvas ZannettouYang ZhangPublished in: CoRR (2023)
Keyphrases
- input image
- image data
- image features
- image collections
- test images
- web images
- image analysis
- original images
- image classification
- image retrieval
- image statistics
- image pixels
- image regions
- random fields
- edge detection
- lighting conditions
- segmentation method
- visual effects
- sample images
- image set
- image structure
- pixel values
- gray value
- illumination conditions
- image content
- region of interest
- imaging process
- three dimensional objects
- text information
- street view
- normalized gradient
- feature points
- image matching
- synthesized images
- set of training images
- image database
- segmentation algorithm
- image segmentation
- image search
- multiscale
- digital imaging
- natural images
- bounding box
- image blocks
- anisotropic diffusion
- object recognition
- high resolution
- textual information
- keypoints
- gaussian noise
- visual features
- image gradient
- complex background
- textual descriptions
- text lines
- secret image
- single image
- image enhancement
- speckle noise
- contrast enhancement
- handwritten documents
- diffusion process
- scanned images
- spatial information
- text graphics