So afaik C3D can split videos and classify them, but most of the datasets, ucf101, for example, contains widely different classes of action. I am looking to a finer detail levels, in the case of football for example, how would one distinguishes a goal shoot and a missed shoot and a free kick etc? C3D can takes me through splitting, but the classification is all messed up. Anyone has any input?