Dataset Link: https://www.kaggle.com/datasets/shawon10/ckplus
Kaggle Notebook: https://www.kaggle.com/code/shasan07/fer-tf-keras
The images were transformed and resized using tf, then we fed them to the pretrained ResNet101 model for fine-Tuning. (Several other popular pre-trained models are also listed along with it).
A learning rate of 1e-4 was used, along with batch size of 8, and the training was conducted for 50 epochs. The best validation accuracy achieved by our model is 99.49%.
Training Summary:
Epoch 1/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 200s 660ms/step - accuracy: 0.6238 - loss: 1.9454 - val_accuracy: 0.5153 - val_loss: 5.1240
Epoch 2/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 23s 231ms/step - accuracy: 0.9347 - loss: 0.2908 - val_accuracy: 0.7092 - val_loss: 1.4095
Epoch 3/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 23s 229ms/step - accuracy: 0.9649 - loss: 0.1546 - val_accuracy: 0.8827 - val_loss: 0.6826
Epoch 4/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 133ms/step - accuracy: 0.9653 - loss: 0.2219 - val_accuracy: 0.7347 - val_loss: 0.9526
Epoch 5/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 133ms/step - accuracy: 0.9696 - loss: 0.2238 - val_accuracy: 0.3010 - val_loss: 85.1607
Epoch 6/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 134ms/step - accuracy: 0.8904 - loss: 0.9186 - val_accuracy: 0.8367 - val_loss: 0.8780
Epoch 7/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 23s 233ms/step - accuracy: 0.9751 - loss: 0.1389 - val_accuracy: 0.9898 - val_loss: 0.0400
Epoch 8/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 133ms/step - accuracy: 0.9860 - loss: 0.0552 - val_accuracy: 0.9694 - val_loss: 0.1258
Epoch 9/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 133ms/step - accuracy: 0.9899 - loss: 0.0893 - val_accuracy: 0.9898 - val_loss: 0.0150
Epoch 10/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 134ms/step - accuracy: 0.9938 - loss: 0.0977 - val_accuracy: 0.9745 - val_loss: 0.1029
Epoch 11/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 133ms/step - accuracy: 0.9665 - loss: 0.1010 - val_accuracy: 0.9541 - val_loss: 0.1482
Epoch 12/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 133ms/step - accuracy: 0.9005 - loss: 0.8510 - val_accuracy: 0.7041 - val_loss: 1.6365
Epoch 13/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 133ms/step - accuracy: 0.9706 - loss: 0.1635 - val_accuracy: 0.7143 - val_loss: 1.2937
Epoch 14/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 23s 233ms/step - accuracy: 0.9956 - loss: 0.0184 - val_accuracy: 0.9949 - val_loss: 0.0115
Epoch 15/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 134ms/step - accuracy: 1.0000 - loss: 7.5090e-04 - val_accuracy: 0.9949 - val_loss: 0.0155
Epoch 16/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 133ms/step - accuracy: 1.0000 - loss: 3.1979e-04 - val_accuracy: 0.9949 - val_loss: 0.0184
Epoch 17/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 132ms/step - accuracy: 1.0000 - loss: 1.1756e-04 - val_accuracy: 0.9898 - val_loss: 0.0195
Epoch 18/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 128ms/step - accuracy: 1.0000 - loss: 6.6640e-05 - val_accuracy: 0.9949 - val_loss: 0.0193
Epoch 19/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 129ms/step - accuracy: 1.0000 - loss: 3.6755e-05 - val_accuracy: 0.9949 - val_loss: 0.0196
Epoch 20/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 131ms/step - accuracy: 1.0000 - loss: 7.6353e-05 - val_accuracy: 0.9898 - val_loss: 0.0208
Epoch 21/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 132ms/step - accuracy: 1.0000 - loss: 6.6136e-05 - val_accuracy: 0.9949 - val_loss: 0.0190
Epoch 22/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 131ms/step - accuracy: 1.0000 - loss: 5.6330e-05 - val_accuracy: 0.9949 - val_loss: 0.0201
Epoch 23/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 132ms/step - accuracy: 1.0000 - loss: 6.6844e-05 - val_accuracy: 0.9949 - val_loss: 0.0206
Epoch 24/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 130ms/step - accuracy: 1.0000 - loss: 2.3923e-05 - val_accuracy: 0.9949 - val_loss: 0.0207
Epoch 25/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 131ms/step - accuracy: 1.0000 - loss: 2.8158e-05 - val_accuracy: 0.9949 - val_loss: 0.0222
Epoch 26/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 131ms/step - accuracy: 1.0000 - loss: 9.8859e-06 - val_accuracy: 0.9949 - val_loss: 0.0224
Epoch 27/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 130ms/step - accuracy: 1.0000 - loss: 2.2506e-05 - val_accuracy: 0.9949 - val_loss: 0.0234
Epoch 28/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 132ms/step - accuracy: 1.0000 - loss: 1.7197e-05 - val_accuracy: 0.9949 - val_loss: 0.0228
Epoch 29/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 132ms/step - accuracy: 1.0000 - loss: 7.2809e-06 - val_accuracy: 0.9949 - val_loss: 0.0227
Epoch 30/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 132ms/step - accuracy: 1.0000 - loss: 7.3699e-06 - val_accuracy: 0.9949 - val_loss: 0.0234
Epoch 31/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 132ms/step - accuracy: 1.0000 - loss: 9.9865e-06 - val_accuracy: 0.9949 - val_loss: 0.0234
Epoch 32/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 130ms/step - accuracy: 1.0000 - loss: 1.5401e-05 - val_accuracy: 0.9949 - val_loss: 0.0239
Epoch 33/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 129ms/step - accuracy: 1.0000 - loss: 9.8768e-06 - val_accuracy: 0.9898 - val_loss: 0.0259
Epoch 34/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 129ms/step - accuracy: 1.0000 - loss: 1.4774e-05 - val_accuracy: 0.9949 - val_loss: 0.0254
Epoch 35/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 129ms/step - accuracy: 1.0000 - loss: 4.6520e-06 - val_accuracy: 0.9949 - val_loss: 0.0239
Epoch 36/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 131ms/step - accuracy: 1.0000 - loss: 5.9569e-06 - val_accuracy: 0.9949 - val_loss: 0.0247
Epoch 37/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 130ms/step - accuracy: 1.0000 - loss: 1.6373e-05 - val_accuracy: 0.9949 - val_loss: 0.0240
Epoch 38/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 132ms/step - accuracy: 0.9871 - loss: 0.1206 - val_accuracy: 0.2245 - val_loss: 15.2200
Epoch 39/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 132ms/step - accuracy: 0.8368 - loss: 1.1002 - val_accuracy: 0.7296 - val_loss: 3.2305
Epoch 40/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 132ms/step - accuracy: 0.8517 - loss: 0.6705 - val_accuracy: 0.8214 - val_loss: 1.8311
Epoch 41/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 131ms/step - accuracy: 0.9806 - loss: 0.1380 - val_accuracy: 0.9745 - val_loss: 0.1141
Epoch 42/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 131ms/step - accuracy: 0.9939 - loss: 0.0226 - val_accuracy: 0.9847 - val_loss: 0.1301
Epoch 43/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 131ms/step - accuracy: 0.9969 - loss: 0.0114 - val_accuracy: 0.9745 - val_loss: 0.2545
Epoch 44/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 131ms/step - accuracy: 0.9973 - loss: 0.0115 - val_accuracy: 0.9898 - val_loss: 0.0900
Epoch 45/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 131ms/step - accuracy: 0.9995 - loss: 0.0010 - val_accuracy: 0.9898 - val_loss: 0.0398
Epoch 46/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 131ms/step - accuracy: 0.9798 - loss: 0.0956 - val_accuracy: 0.9796 - val_loss: 0.1385
Epoch 47/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 131ms/step - accuracy: 0.9904 - loss: 0.0248 - val_accuracy: 0.9898 - val_loss: 0.0350
Epoch 48/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 129ms/step - accuracy: 0.9945 - loss: 0.0149 - val_accuracy: 0.9898 - val_loss: 0.0318
Epoch 49/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 129ms/step - accuracy: 1.0000 - loss: 2.7206e-04 - val_accuracy: 0.9949 - val_loss: 0.0325
Epoch 50/50
99/99 ━━━━━━━━━━━━━━━━━━━━ 13s 130ms/step - accuracy: 1.0000 - loss: 7.0056e-04 - val_accuracy: 0.9949 - val_loss: 0.0352
Obtained Results:
Classification Report:
precision recall f1-score support
Anger 1.00 1.00 1.00 31
Contempt 1.00 1.00 1.00 10
Disgust 1.00 1.00 1.00 36
Fear 1.00 1.00 1.00 20
Happy 1.00 1.00 1.00 43
Sad 0.88 1.00 0.93 7
Surprise 1.00 0.98 0.99 49
accuracy 0.99 196
macro avg 0.98 1.00 0.99 196
weighted avg 1.00 0.99 1.00 196
