This dataset contains BIO tagged utterances of image edit requests in imperative low-level form. The edit type is "adjust", and contains 3 entity types (1) referring expression (2) attribute (3) value. (1) referring expression is an phrase referring to an object in an image. (2) attribute is among (a) brightness (b) contrast (c) hue (d) saturation (e) lightness. (3) value is between -100 and 100. It has also been split into train/dev/test, each with 2046/243/241 utterances respectively. There is also a preprocessed version ".abstract.txt", where (2) attribute and (3) value are replaced into "<attribute>" and "<value>" tokens (delexicalized).