Benchmark
JEdit-1M
A large-scale evaluation dataset for image editing tasks in joint text-and-image generation, containing one million samples. It is used to assess joint accuracy between text and image.
Timeline (last 90 days)
Sep 30
The team released three datasets: JEdit-1M (million-scale image editing), JMaze-200K (approximately 200,000 maze samples), and JNono-200K (approximately 200,000 nonogram samples).
Unverified50%