💡📊Training data used in ComCLIP
CLIP (Contrastive Language-Image Pre-Training) is a neural network developed by OpenAI to perform visual as well as language comprehension tasks. The algorithms aim to understand the relationship between text and images.
ComCLIP is an improved version of CLIP for matching text and graphic representations. ComCLIP can mitigate spurious correlations introduced by pre-trained CLIP models and dynamically estimate the importance of each component. Experiments were conducted on four datasets for compositional matching between images and texts.
These datasets are publicly available on the Internet and can be found at this link