Conducted Experiments using 5 models on Food 101 Dataset
Experiment 1: Model_0
->Data: 10 Classes of Food 101 Data (10% Training Data Only)
->Pre processing: None
->Model_Architecture: Feature Extraction -> EfficientNet-B0 (Pretrained on ImageNet, all layers Frozen with no top)
Accuracy: 82.68%, Loss: 0.6472
Experiment 2 (Changes: Data-> 1% and Pre_Processing): Model_1
->Data: 10 Classes of Food 101 Data (1% Training Data Only)
->Pre processing: Data Augmentation, Random Flips, Rotation, Zoom, Height, Width
->Model_Architecture: Feature Extraction -> EfficientNet-B0 (Pretrained on ImageNet, all layers Frozen with no top)
Accuracy: 35.21%, Loss: 1.8930
Experiment 3 (Changes: Data-> 10%): Model_2
->Data: 10 Classes of Food 101 Data (10% Training Data Only)
->Pre processing: Data Augmentation, Random Flips, Rotation, Zoom, Height, Width
->Model_Architecture: Feature Extraction -> EfficientNet-B0 (Pretrained on ImageNet, all layers Frozen with no top)
Accuracy: 51.24, Loss: 1.8918
Experiment 4 (Changes: Model_Architecture): Model_3
->Data: 10 Classes of Food 101 Data (10% Training Data Only)
->Pre processing: Data Augmentation, Random Flips, Rotation, Zoom, Height, Width
->Model_Architecture: Fine Tuning -> EfficientNet-B0 with top layer trained on custom data, top 10 layer unfrozen
Accuracy: 48.74%, Loss: 2.0333
Experiment 5: Model_4 (Changes: Data)
-> Data: 10 Classes of Food 101 Data (100% Training Data)
->Pre processing: Data Augmentation, Random Flips, Rotation, Zoom, Height, Width
->Model_Architecture: Fine Tuning -> EfficientNet-B0 with top layer trained on custom data, top 10 layer unfrozen
Accuracy: 84.10 , Loss: 0.4571
