다음은 모델 학습에 대한 결과입니다.
Epoch: 01 | Epoch Time: 4m 4s Train Loss: 0.698 | Train Acc @1: 50.24% | Train Acc @5: 100.00% Valid Loss: 0.699 | Valid Acc @1: 51.19% | Valid Acc @5: 100.00% Epoch: 02 | Epoch Time: 4m 10s Train Loss: 0.693 | Train Acc @1: 50.48% | Train Acc @5: 100.00% Valid Loss: 0.699 | Valid Acc @1: 51.19% | Valid Acc @5: 100.00% Epoch: 03 | Epoch Time: 4m 13s Train Loss: 0.701 | Train Acc @1: 49.04% | Train Acc @5: 100.00% Valid Loss: 0.696 | Valid Acc @1: 47.17% | Valid Acc @5: 100.00% Epoch: 04 | Epoch Time: 4m 11s Train Loss: 0.696 | Train Acc @1: 49.52% | Train Acc @5: 100.00% Valid Loss: 0.698 | Valid Acc @1: 48.96% | Valid Acc @5: 100.00% Epoch: 05 | Epoch Time: 4m 2s Train Loss: 0.695 | Train Acc @1: 49.52% | Train Acc @5: 100.00% Valid Loss: 0.700 | Valid Acc @1: 50.00% | Valid Acc @5: 100.00% Epoch: 06 | Epoch Time: 4m 3s Train Loss: 0.686 | Train Acc @1: 58.41% | Train Acc @5: 100.00% Valid Loss: 0.693 | Valid Acc @1: 49.85% | Valid Acc @5: 100.00% Epoch: 07 | Epoch Time: 4m 7s Train Loss: 0.686 | Train Acc @1: 52.64% | Train Acc @5: 100.00% Valid Loss: 0.690 | Valid Acc @1: 51.93% | Valid Acc @5: 100.00% Epoch: 08 | Epoch Time: 4m 7s Train Loss: 0.688 | Train Acc @1: 57.45% | Train Acc @5: 100.00% Valid Loss: 0.693 | Valid Acc @1: 49.55% | Valid Acc @5: 100.00% Epoch: 09 | Epoch Time: 4m 0s Train Loss: 0.690 | Train Acc @1: 55.29% | Train Acc @5: 100.00% Valid Loss: 0.692 | Valid Acc @1: 54.32% | Valid Acc @5: 100.00% Epoch: 10 | Epoch Time: 4m 1s Train Loss: 0.690 | Train Acc @1: 57.69% | Train Acc @5: 100.00% Valid Loss: 0.693 | Valid Acc @1: 51.93% | Valid Acc @5: 100.00%
역시 모델 학습 결과에 대한 오차와 정확도 측면에서 성능이 좋지 않습니다. 계속 언급하지만 이미지 데이터를 늘린다면 성능은 좋아질 수 있습니다. 이 예제의 목적은 성능 향상이 아닌 CNN 관련 네트워크의 사용 방법이므로, 빠른 학습을 위해 데이터 개수를 제한시켰기 때문에 성능은 좋지 않습니다.