How We're Using CLIP
Go Deeper!
CLIP: Trained on 400M image/text pairs, OpenAI, 2021
text_embedding: how CLIP maps the paint.wtf prompt into its feature space
Image_embedding: how CLIP maps the user submission into its feature space
Winning paint.wtf: minimizing distance between the prompt and the user drawing
text_embedding: how CLIP maps the paint.wtf prompt into its feature space
Image_embedding: how CLIP maps the user submission into its feature space
Winning paint.wtf: minimizing distance between the prompt and the user drawing
Paint.wtf Scoring
CLIP's Interpretation of the Paint.wtf Prompt
CLIP's Interpretation of User-Submitted Drawing
Cosine Similarity
CLIP's Interpretation of the Paint.wtf Prompt
CLIP's Interpretation of User-Submitted Drawing
Cosine Similarity