EVALALIGN: Supervised Fine-Tuning Multimodal LLMs with Human-Aligned Data for Evaluating Text-to-Image Models.
Zhiyu TanXiaomeng YangLuozheng QinMengping YangCheng ZhangHao LiPublished in: CoRR (2024)
Keyphrases
- image data
- fine tuning
- data sets
- experimental data
- database
- historical data
- data analysis
- data structure
- single image
- image classification
- image features
- feature selection
- high quality
- low level
- multiscale
- data sources
- data points
- high resolution
- prior knowledge
- input image
- markov random field
- input data
- image representation
- keywords
- segmentation method
- image segmentation
- statistical methods
- viable alternative
- human subjects
- image collections
- knowledge discovery
- training data