Is end-to-end learning enough for fitness activity recognition?