Released by Google with their OpenImages dataset, this annotation format is very similar to YOLO Darknet except it contains several other columns and, instead of a numeric class ID, the class ID is a string of the form /m/xxxxx.
No known models train using the OpenImages annotation format but Roboflow can import these annotations so you can convert them into whatever other format you need!
Below, learn the structure of OpenImages CSV.
ImageID,Source,LabelName,Confidence,XMin,XMax,YMin,YMax,IsOccluded,IsTruncated,IsGroupOf,IsDepiction,IsInside
000026e7ee790996,xclick,/m/07j7r,1,0.071875,0.1453125,0.20625,0.39166668,0,1,1,0,0
000026e7ee790996,xclick,/m/07j7r,1,0.4390625,0.571875,0.26458332,0.43541667,0,1,1,0,0
000026e7ee790996,xclick,/m/07j7r,1,0.66875,1,0,0.5520833,0,1,1,0,0
000062a39995e348,xclick,/m/015p6,1,0.20620842,0.849224,0.15463917,1,0,0,0,0,0
000062a39995e348,xclick,/m/05s2s,1,0.13747229,0.37694013,0,0.88365245,1,1,0,0,0
0000c64e1253d68f,xclick,/m/0k4j,1,0,0.5132743,0.32079646,0.6902655,0,1,0,0,0
0000c64e1253d68f,xclick,/m/0k4j,1,0.016224189,0.26843658,0.29867256,0.46238938,1,0,0,0,0
0000c64e1253d68f,xclick,/m/0k4j,1,0.48082596,0.9041298,0.23230088,0.48893806,1,0,0,0,0
0000c64e1253d68f,xclick,/m/0k4j,1,0.7522124,1,0.3318584,0.65707964,0,1,0,0,0With Roboflow supervision, an open source Python package with utilities for completing computer vision tasks, you can merge and split detections in OpenImages CSV. Read our dedicated guides to learn how to merge and split OpenImages CSV detections.
Below, see model architectures that require data in the OpenImages CSV format when training a new model.
On each page below, you can find links to our guides that show how to plot predictions from the model, and complete other common tasks like detecting small objects with the model.