The PIPE (Paint by InPaint Edit) dataset is designed to enhance the efficacy of mask-free, instruction-following image editing models by providing a large-scale collection of image pairs and diverse object addition instructions. Comprising approximately 1 million image pairs, PIPE includes both source and target images, along with corresponding natural language instructions for object addition. The dataset leverages extensive… See the full description on the dataset page: https://huggingface.co/datasets/paint-by-inpaint/PIPE.
