Results on Caltech-101 so far (30 learning examples per category), confusion matrix:
The average recognition rate is 46%, which is remarkable for
such a simple method. Categories are in order:
class: 1/102 'BACKGROUND_Google', rate= 2.069%, n= 435 class: 2/102 ' Faces', rate= 60.000%, n= 405 class: 3/102 ' Faces_easy', rate= 94.321%, n= 405 class: 4/102 ' Leopards', rate= 32.353%, n= 170 class: 5/102 ' Motorbikes', rate= 92.318%, n= 768 class: 6/102 ' accordion', rate=100.000%, n= 25 class: 7/102 ' airplanes', rate= 90.260%, n= 770 class: 8/102 ' anchor', rate= 25.000%, n= 12 class: 9/102 ' ant', rate= 25.000%, n= 12 class: 10/102 ' barrel', rate= 5.882%, n= 17 class: 11/102 ' bass', rate= 8.333%, n= 24 class: 12/102 ' beaver', rate= 0.000%, n= 16 class: 13/102 ' binocular', rate= 33.333%, n= 3 class: 14/102 ' bonsai', rate= 28.571%, n= 98 class: 15/102 ' brain', rate= 54.412%, n= 68 class: 16/102 ' brontosaurus', rate= 46.154%, n= 13 class: 17/102 ' buddha', rate= 32.727%, n= 55 class: 18/102 ' butterfly', rate= 9.836%, n= 61 class: 19/102 ' camera', rate= 45.000%, n= 20 class: 20/102 ' cannon', rate= 30.769%, n= 13 class: 21/102 ' car_side', rate= 79.570%, n= 93 class: 22/102 ' ceiling_fan', rate= 29.412%, n= 17 class: 23/102 ' cellphone', rate= 68.966%, n= 29 class: 24/102 ' chair', rate= 12.500%, n= 32 class: 25/102 ' chandelier', rate= 29.870%, n= 77 class: 26/102 ' cougar_body', rate= 11.765%, n= 17 class: 27/102 ' cougar_face', rate= 25.641%, n= 39 class: 28/102 ' crab', rate= 6.977%, n= 43 class: 29/102 ' crayfish', rate= 22.500%, n= 40 class: 30/102 ' crocodile', rate= 20.000%, n= 20 class: 31/102 ' crocodile_head', rate= 9.524%, n= 21 class: 32/102 ' cup', rate= 29.630%, n= 27 class: 33/102 ' dalmatian', rate= 54.054%, n= 37 class: 34/102 ' dollar_bill', rate= 95.455%, n= 22 class: 35/102 ' dolphin', rate= 20.000%, n= 35 class: 36/102 ' dragonfly', rate= 47.368%, n= 38 class: 37/102 ' electric_guitar', rate= 24.444%, n= 45 class: 38/102 ' elephant', rate= 29.412%, n= 34 class: 39/102 ' emu', rate= 26.087%, n= 23 class: 40/102 ' euphonium', rate= 64.706%, n= 34 class: 41/102 ' ewer', rate= 20.000%, n= 55 class: 42/102 ' ferry', rate= 48.649%, n= 37 class: 43/102 ' flamingo', rate= 21.622%, n= 37 class: 44/102 ' flamingo_head', rate= 46.667%, n= 15 class: 45/102 ' garfield', rate=100.000%, n= 4 class: 46/102 ' gerenuk', rate= 25.000%, n= 4 class: 47/102 ' gramophone', rate= 14.286%, n= 21 class: 48/102 ' grand_piano', rate= 82.609%, n= 69 class: 49/102 ' hawksbill', rate= 47.143%, n= 70 class: 50/102 ' headphone', rate= 83.333%, n= 12 class: 51/102 ' hedgehog', rate= 16.667%, n= 24 class: 52/102 ' helicopter', rate= 34.483%, n= 58 class: 53/102 ' ibis', rate= 22.000%, n= 50 class: 54/102 ' inline_skate', rate=100.000%, n= 1 class: 55/102 ' joshua_tree', rate= 55.882%, n= 34 class: 56/102 ' kangaroo', rate= 17.857%, n= 56 class: 57/102 ' ketch', rate= 36.905%, n= 84 class: 58/102 ' lamp', rate= 19.355%, n= 31 class: 59/102 ' laptop', rate= 43.137%, n= 51 class: 60/102 ' llama', rate= 10.417%, n= 48 class: 61/102 ' lobster', rate= 18.182%, n= 11 class: 62/102 ' lotus', rate= 30.556%, n= 36 class: 63/102 ' mandolin', rate= 46.154%, n= 13 class: 64/102 ' mayfly', rate= 10.000%, n= 10 class: 65/102 ' menorah', rate= 70.175%, n= 57 class: 66/102 ' metronome', rate=100.000%, n= 2 class: 67/102 ' minaret', rate=100.000%, n= 46 class: 68/102 ' nautilus', rate= 32.000%, n= 25 class: 69/102 ' octopus', rate= 20.000%, n= 5 class: 70/102 ' okapi', rate= 66.667%, n= 9 class: 71/102 ' pagoda', rate=100.000%, n= 17 class: 72/102 ' panda', rate= 12.500%, n= 8 class: 73/102 ' pigeon', rate= 33.333%, n= 15 class: 74/102 ' pizza', rate= 17.391%, n= 23 class: 75/102 ' platypus', rate= 75.000%, n= 4 class: 76/102 ' pyramid', rate= 40.741%, n= 27 class: 77/102 ' revolver', rate= 78.846%, n= 52 class: 78/102 ' rhino', rate= 51.724%, n= 29 class: 79/102 ' rooster', rate= 68.421%, n= 19 class: 80/102 ' saxophone', rate= 50.000%, n= 10 class: 81/102 ' schooner', rate= 57.576%, n= 33 class: 82/102 ' scissors', rate= 77.778%, n= 9 class: 83/102 ' scorpion', rate= 35.185%, n= 54 class: 84/102 ' sea_horse', rate= 40.741%, n= 27 class: 85/102 ' snoopy', rate=100.000%, n= 5 class: 86/102 ' soccer_ball', rate= 58.824%, n= 34 class: 87/102 ' stapler', rate= 46.667%, n= 15 class: 88/102 ' starfish', rate= 26.786%, n= 56 class: 89/102 ' stegosaurus', rate= 55.172%, n= 29 class: 90/102 ' stop_sign', rate= 97.059%, n= 34 class: 91/102 ' strawberry', rate= 20.000%, n= 5 class: 92/102 ' sunflower', rate= 74.545%, n= 55 class: 93/102 ' tick', rate= 84.211%, n= 19 class: 94/102 ' trilobite', rate= 87.500%, n= 56 class: 95/102 ' umbrella', rate= 46.667%, n= 45 class: 96/102 ' watch', rate= 51.196%, n= 209 class: 97/102 ' water_lilly', rate= 14.286%, n= 7 class: 98/102 ' wheelchair', rate= 41.379%, n= 29 class: 99/102 ' wild_cat', rate= 0.000%, n= 4 class: 100/102 ' windsor_chair', rate= 84.615%, n= 26 class: 101/102 ' wrench', rate=100.000%, n= 9 class: 102/102 ' yin_yang', rate= 90.000%, n= 30 class: *** final score: 45.903 ***
State of the art
For comparison, this [1] is the state of the art for Caltech-101 (67% average recognition rate):
[1] H. Zhang, A. C. Berg, M. Maire, and J. Malik, “SVM-KNN: Discriminative nearest neighbor classification for visual category recognition,” in Proc. CVPR, 2006.