The dataset can be found here http://crcv.ucf.edu/data/UCF101.php
The model requires about 15-18 hours to train on GPU
The full output is given in the output.out file.
('Fencing', 0.97058821, 34.0) ('Kayaking', 0.97222221, 36.0) ('Rowing', 0.97222221, 36.0) ('ParallelBars', 0.97297299, 37.0) ('Biking', 0.97368419, 38.0) ('CliffDiving', 0.97435898, 39.0) ('PoleVault', 0.97500002, 40.0) ('IceDancing', 0.97826087, 46.0) ('HorseRiding', 0.97959185, 49.0) ('CuttingInKitchen', 1.0, 33.0) ('Billiards', 1.0, 40.0) ('PlayingDhol', 1.0, 49.0) ('TrampolineJumping', 1.0, 32.0) ('Diving', 1.0, 45.0) ('PlayingPiano', 1.0, 28.0) ('Haircut', 1.0, 33.0) ('BasketballDunk', 1.0, 37.0) ('SumoWrestling', 1.0, 34.0) ('SoccerPenalty', 1.0, 41.0) ('HorseRace', 1.0, 35.0) ('SkyDiving', 1.0, 31.0) ('Skijet', 1.0, 28.0) ('WritingOnBoard', 1.0, 45.0) ('PlayingSitar', 1.0, 44.0) ('TableTennisShot', 1.0, 39.0)
The Resnet code has been borrowed from https://github.com/ry/tensorflow-resnet