We are using traditional categorization accuracy as our main metric. The  participants just need to predict the labels of image and save them in a csv  file. The sample format of the submission file is available in "Data".