Tf Image Classifier

• Overview - Model Evaluation & Deployment

Model Evaluation

Keras Applications are deep learning models that are made available alongside pre-trained weights. These models can be used for prediction, feature extraction, and fine-tuning.

Weights are downloaded automatically when instantiating a model. They are stored at ~/.keras/models/:

Available Models - Excerp

Model	Size (MB)	Top-1 Accuracy	Top-5 Accuracy	Parameters	Depth	Time (ms) per inference step (CPU)	Time (ms) per inference step (GPU)
Xception	88	79.0%	94.5%	22.9M	81	109.4	8.1
ResNet50V2	98	76.0%	93.0%	25.6M	103	45.6	4.4
InceptionV3	92	77.9%	93.7%	23.9M	189	42.2	6.9
MobileNet	16	70.4%	89.5%	4.3M	55	22.6	3.4
MobileNetV2	14	71.3%	90.1%	3.5M	105	25.9	3.8
NASNetMobile	23	74.4%	91.9%	5.3M	389	27.0	6.7
EfficientNetV2B0	29	78.7%	94.3%	7.2M	-	-	-
EfficientNetV2B1	34	79.8%	95.0%	8.2M	-	-	-
EfficientNetV2B2	42	80.5%	95.1%	10.2M	-	-	-
EfficientNetV2B3	59	82.0%	95.8%	14.5M	-	-	-
EfficientNetV2S	88	83.9%	96.7%	21.6M	-	-	-
EfficientNetV2M	220	85.3%	97.4%	54.4M	-	-	-
EfficientNetV2L	479	85.7%	97.5%	119.0M	-	-	-

Used Models

• EfficientNetV2B0
• EfficientNetV2S
• Xception
• InceptionV3
• NASNetMobile
• MobileNetV3Small
• MobileNetV3Large
• MobileNetV2
• vit-base-patch16-224
• DeiT

import matplotlib.pyplot as plt
import pandas as pd
import seaborn as sns

Model Metrics

model_metrics = [
    [ 'MobileNetV3Small', 0.3474, 0.9178, 16.753, 75, 260, 7, 76 ],
    [ 'MobileNetV3Large', 0.0658, 0.9930, 34.593, 58, 206, 5, 40 ],
    [ 'MobileNetV2', 1.2054, 0.6573, 47.857, 96, 331, 8, 79 ],
    [ 'EfficientNetV2B0', 0.3779, 0.9084, 45.989, 47, 164, 8, 68 ],
    [ 'EfficientNetV2S', 0.3431, 0.9013, 107.201, 66, 230, 21, 178 ],
    [ 'Xception', 0.6363, 0.8188, 110.849, 69, 240, 18, 173 ],
    [ 'InceptionV3', 2.7047, 0.3087, 116.969, 48, 167, 10, 85 ],
    [ 'NASNetMobile', 3.2046, 0.1838, 38.513, 247, 857, 41, 423 ],
    [ 'ViT', 0.5979, 0.8589, 350.537, 172, 597, 21, 203 ],
    [ 'DeiT', 0.7199, 0.7900, 350.737, 174, 604, 21, 204 ]
]

data_columns = [
    'Model',
    'Loss',
    'Accuracy',
    'Tf Model Size [MB]',
    'Training Training Set [s]',
    'Training [ms/step]',
    'Prediction Validation Set [s]',
    'Prediction [ms/step]'
]
model_metrics_df = pd.DataFrame(model_metrics, columns=data_columns)
model_metrics_df = model_metrics_df.set_index('Model')



model_metrics_df

Model	Loss	Accuracy	Tf Model Size [MB]	Training Training Set [s]	Training [ms/step]	Prediction Validation Set [s]	Prediction [ms/step]
MobileNetV3Small	0.3474	0.9178	16.753	75	260	7	76
MobileNetV3Large	0.0658	0.9930	34.593	58	206	5	40
MobileNetV2	1.2054	0.6573	47.857	96	331	8	79
EfficientNetV2B0	0.3779	0.9084	45.989	47	164	8	68
EfficientNetV2S	0.3431	0.9013	107.201	66	230	21	178
Xception	0.6363	0.8188	110.849	69	240	18	173
InceptionV3	2.7047	0.3087	116.969	48	167	10	85
NASNetMobile	3.2046	0.1838	38.513	247	857	41	423
ViT	0.5979	0.8589	350.537	172	597	21	203
DeiT	0.7199	0.7900	350.737	174	604	21	204

plt.figure(figsize=(12, 5))

sns.set(style='darkgrid')

sns.scatterplot(
    x='Tf Model Size [MB]',
    y='Accuracy',
    data=model_metrics_df,
    s=100,
    alpha=1,
    hue='Model',
    palette='nipy_spectral'
).set_title('Accuracy vs Model Size')

plt.savefig('./assets/Model_Eval_04.webp', bbox_inches='tight')

plt.figure(figsize=(12, 5))

sns.set(style='darkgrid')

sns.scatterplot(
    x='Tf Model Size [MB]',
    y='Loss',
    data=model_metrics_df,
    s=100,
    alpha=1,
    hue='Model',
    palette='nipy_spectral'
).set_title('Loss vs Model Size')

plt.savefig('./assets/Model_Eval_05.webp', bbox_inches='tight')

plt.figure(figsize=(12, 5))

sns.set(style='darkgrid')

sns.scatterplot(
    x='Training [ms/step]',
    y='Accuracy',
    data=model_metrics_df,
    s=100,
    alpha=1,
    hue='Model',
    palette='nipy_spectral'
).set_title('Accuracy vs Training Time')

plt.savefig('./assets/Model_Eval_06.webp', bbox_inches='tight')

plt.figure(figsize=(12, 5))

sns.set(style='darkgrid')

sns.scatterplot(
    x='Prediction [ms/step]',
    y='Accuracy',
    data=model_metrics_df,
    s=100,
    alpha=1,
    hue='Model',
    palette='nipy_spectral'
).set_title('Loss vs Prediction Time')

plt.savefig('./assets/Model_Eval_07.webp', bbox_inches='tight')

Loss Curves

loss_curves = [
    [ 'MobileNetV3Small', './assets/MobileNetV3Small_FT_01.webp' ],
    [ 'MobileNetV3Large', './assets/MobileNetV3Large_FT_01.webp' ],
    [ 'EfficientNetV2B0', './assets/EfficientNetV2B0_01.webp' ],
    [ 'EfficientNetV2S', './assets/EfficientNetV2S_01.webp' ],
    [ 'Xception', './assets/Xception_01.webp' ],
    [ 'InceptionV3', './assets/InceptionV3_FT_01.webp' ],
    [ 'NASNetMobile', './assets/NASNetMobile_FT_01.webp' ],
    [ 'MobileNetV2', './assets/MobileNetV2_FT_01.webp' ],
    [ 'ViT', './assets/ViT_01.webp' ],
    [ 'DeiT', './assets/DeiT_01.webp' ]
]

plt.figure(figsize=(16,5))

for model, curve in loss_curves:
    for i in range(len(loss_curves)):
        ax = plt.subplot(2,5,i+1)
        plt.title(loss_curves[i][0])
        image = plt.imread(loss_curves[i][1])
        plt.imshow(image)
        plt.axis('off')
        
plt.savefig('./assets/Model_Eval_01.webp', bbox_inches='tight')

Accuracy Curves

accuracy_curves = [
    [ 'MobileNetV3Small', './assets/MobileNetV3Small_FT_02.webp' ],
    [ 'MobileNetV3Large', './assets/MobileNetV3Large_FT_02.webp' ],
    [ 'EfficientNetV2B0', './assets/EfficientNetV2B0_02.webp' ],
    [ 'EfficientNetV2S', './assets/EfficientNetV2S_02.webp' ],
    [ 'Xception', './assets/Xception_02.webp' ],
    [ 'InceptionV3', './assets/InceptionV3_FT_02.webp' ],
    [ 'NASNetMobile', './assets/NASNetMobile_FT_02.webp' ],
    [ 'MobileNetV2', './assets/MobileNetV2_FT_02.webp' ],
    [ 'ViT', './assets/ViT_02.webp' ],
    [ 'DeiT', './assets/DeiT_02.webp' ]
]

plt.figure(figsize=(16,5))

for model, curve in accuracy_curves:
    for i in range(len(accuracy_curves)):
        ax = plt.subplot(2,5,i+1)
        plt.title(accuracy_curves[i][0])
        image = plt.imread(accuracy_curves[i][1])
        plt.imshow(image)
        plt.axis('off')
        
plt.savefig('./assets/Model_Eval_02.webp', bbox_inches='tight')

Confusion

confusion_matrices = [
    [ 'MobileNetV3Small', './assets/MobileNetV3Small_FT_04.webp' ],
    [ 'MobileNetV3Large', './assets/MobileNetV3Large_FT_04.webp' ],
    [ 'EfficientNetV2B0', './assets/EfficientNetV2B0_04.webp' ],
    [ 'EfficientNetV2S', './assets/EfficientNetV2S_04.webp' ],
    [ 'Xception', './assets/Xception_FT_04.webp' ],
    [ 'NASNetMobile', './assets/NASNetMobile_FT_04.webp' ],
    [ 'MobileNetV2', './assets/MobileNetV2_FT_04.webp' ],
    [ 'ViT', './assets/ViT_04.webp' ],
    [ 'DeiT', './assets/DeiT_04.webp' ]
]

plt.figure(figsize=(14,12))

for model, curve in confusion_matrices:
    for i in range(len(confusion_matrices)):
        ax = plt.subplot(3,3,i+1)
        plt.title(confusion_matrices[i][0])
        image = plt.imread(confusion_matrices[i][1])
        plt.imshow(image)
        plt.axis('off')
        
plt.savefig('./assets/Model_Eval_03.webp', bbox_inches='tight')

Model Prediction

MobileNetV3Small

MobileNetV3Small_Predictions = [
    [ './assets/MobileNetV3Small_Prediction_01.webp' ],
    [ './assets/MobileNetV3Small_Prediction_02.webp' ],
    [ './assets/MobileNetV3Small_Prediction_03.webp' ]
]

plt.figure(figsize=(16,8))

for prediction in MobileNetV3Small_Predictions:
    for i in range(len(MobileNetV3Small_Predictions)):
        ax = plt.subplot(1,3,i+1)
        image = plt.imread(MobileNetV3Small_Predictions[i][0])
        plt.imshow(image)
        plt.axis('off')
        
plt.savefig('./assets/Model_Eval_MobileNetV3Small_01.webp', bbox_inches='tight')

MobileNetV2

MobileNetV2_Predictions = [
    [ './assets/MobileNetV2_Prediction_01.webp' ],
    [ './assets/MobileNetV2_Prediction_02.webp' ],
    [ './assets/MobileNetV2_Prediction_03.webp' ]
]

plt.figure(figsize=(16,8))

for prediction in MobileNetV2_Predictions:
    for i in range(len(MobileNetV2_Predictions)):
        ax = plt.subplot(1,3,i+1)
        image = plt.imread(MobileNetV2_Predictions[i][0])
        plt.imshow(image)
        plt.axis('off')
        
plt.savefig('./assets/Model_Eval_MobileNetV2_01.webp', bbox_inches='tight')

MobileNetV3Large

MobileNetV3Large_Predictions = [
    [ './assets/MobileNetV3Large_Prediction_01.webp' ],
    [ './assets/MobileNetV3Large_Prediction_02.webp' ],
    [ './assets/MobileNetV3Large_Prediction_03.webp' ]
]

plt.figure(figsize=(16,8))

for prediction in MobileNetV3Small_Predictions:
    for i in range(len(MobileNetV3Small_Predictions)):
        ax = plt.subplot(1,3,i+1)
        image = plt.imread(MobileNetV3Small_Predictions[i][0])
        plt.imshow(image)
        plt.axis('off')
        
plt.savefig('./assets/Model_Eval_MobileNetV3Large_01.webp', bbox_inches='tight')

EfficientNetV2B0

MobileNetV3Small_Predictions = [
    [ './assets/EfficientNetV2B0_Prediction_01.webp' ],
    [ './assets/EfficientNetV2B0_Prediction_02.webp' ],
    [ './assets/EfficientNetV2B0_Prediction_03.webp' ]
]

plt.figure(figsize=(16,8))

for prediction in MobileNetV3Small_Predictions:
    for i in range(len(MobileNetV3Small_Predictions)):
        ax = plt.subplot(1,3,i+1)
        image = plt.imread(MobileNetV3Small_Predictions[i][0])
        plt.imshow(image)
        plt.axis('off')
        
plt.savefig('./assets/Model_Eval_EfficientNetV2B0_01.webp', bbox_inches='tight')

EfficientNetV2S

EfficientNetV2S_Predictions = [
    [ './assets/EfficientNetV2S_Prediction_01.webp' ],
    [ './assets/EfficientNetV2S_Prediction_02.webp' ],
    [ './assets/EfficientNetV2S_Prediction_03.webp' ]
]

plt.figure(figsize=(16,8))

for prediction in EfficientNetV2S_Predictions:
    for i in range(len(EfficientNetV2S_Predictions)):
        ax = plt.subplot(1,3,i+1)
        image = plt.imread(EfficientNetV2S_Predictions[i][0])
        plt.imshow(image)
        plt.axis('off')
        
plt.savefig('./assets/Model_Eval_EfficientNetV2S_01.webp', bbox_inches='tight')

Xception

Xception_Predictions = [
    [ './assets/Xception_Prediction_01.webp' ],
    [ './assets/Xception_Prediction_02.webp' ],
    [ './assets/Xception_Prediction_03.webp' ]
]

plt.figure(figsize=(16,8))

for prediction in Xception_Predictions:
    for i in range(len(Xception_Predictions)):
        ax = plt.subplot(1,3,i+1)
        image = plt.imread(Xception_Predictions[i][0])
        plt.imshow(image)
        plt.axis('off')
        
plt.savefig('./assets/Model_Eval_Xception_01.webp', bbox_inches='tight')

NASNetMobile

NASNetMobile_Predictions = [
    [ './assets/NASNetMobile_Prediction_01.webp' ],
    [ './assets/NASNetMobile_Prediction_02.webp' ],
    [ './assets/NASNetMobile_Prediction_03.webp' ]
]

plt.figure(figsize=(16,8))

for prediction in NASNetMobile_Predictions:
    for i in range(len(NASNetMobile_Predictions)):
        ax = plt.subplot(1,3,i+1)
        image = plt.imread(NASNetMobile_Predictions[i][0])
        plt.imshow(image)
        plt.axis('off')
        
plt.savefig('./assets/Model_Eval_NASNetMobile_01.webp', bbox_inches='tight')

ViT

ViT_Predictions = [
    [ './assets/ViT_Prediction_01.webp' ],
    [ './assets/ViT_Prediction_02.webp' ],
    [ './assets/ViT_Prediction_03.webp' ]
]

plt.figure(figsize=(16,8))

for prediction in ViT_Predictions:
    for i in range(len(ViT_Predictions)):
        ax = plt.subplot(1,3,i+1)
        image = plt.imread(ViT_Predictions[i][0])
        plt.imshow(image)
        plt.axis('off')
        
plt.savefig('./assets/Model_Eval_ViT_01.webp', bbox_inches='tight')

DeiT

DeiT_Predictions = [
    [ './assets/DeiT_Prediction_01.webp' ],
    [ './assets/DeiT_Prediction_02.webp' ],
    [ './assets/DeiT_Prediction_03.webp' ]
]

plt.figure(figsize=(16,8))

for prediction in DeiT_Predictions:
    for i in range(len(DeiT_Predictions)):
        ax = plt.subplot(1,3,i+1)
        image = plt.imread(DeiT_Predictions[i][0])
        plt.imshow(image)
        plt.axis('off')
        
plt.savefig('./assets/Model_Eval_DeiT_01.webp', bbox_inches='tight')