Bias bugs

COCO with gender:

bugs: given horst, surfboard, skis, or frisbee and woman images, COCO with gender model predicts man

imSitu:

bugs: given inside or room and man images, imSitu model predicts woman

bugs: given outside and woman images, imSitu model predicts man

ImageNet:

bugs: given Australian_terrier images, ImageNet model predicts Norwich_terrier. The model tends to be more likely to make mistake betwen Australian_terrier and Norwich_terrier, compared with rubber_eraser and Norwich_terrier.

bugs: given langur images, ImageNet model predicts macaque. The model tends to be more likely to make mistake betwen langur and macaque, compared with screwdriver and macaque.

bugs: given Siberian_husky images, ImageNet model predicts Eskimo_dog. The model tends to be more likely to make mistake betwen Eskimo_dog and Siberian_husky, compared with rapeseed and Siberian_husky.

bugs: given night_snake images, ImageNet model predicts king_snake. The model tends to be more likely to make mistake betwen night_snake and king_snake, compared with beacon and king_snake.

bugs: given spider_monkey images, ImageNet model predicts Titi. The model tends to be more likely to make mistake betwen Titi and spider_monkey, compared with wooden_spoon and spider_monkey.