Cat - Dog Classifier

Clone Repo

git clone https://github.com/shoryasethia/Cats-Dog-InceptionV3
cd Cats-Dog-InceptionV3

Dataset

• Download it from here

Transfer Learning

• Used InceptionV3 as base model

from tensorflow.keras.applications.inception_v3 import InceptionV3

pre_trained_model = InceptionV3(input_shape = (150, 150, 3),
                                include_top = False,              
                                weights = None)           

• Used weights of Imagnet dataset, downloaded from here

pre_trained_weights_file = 'inception_v3_weights_tf_dim_ordering_tf_kernels_notop.h5'

pre_trained_model.load_weights(pre_trained_weights_file)

• Freezed those weights and removed/discarded last few layers

pre_trained_model.trainable = False   

• Added following layers further

Model: "sequential_1"
┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓
┃ Layer (type)                    ┃ Output Shape           ┃       Param # ┃
┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩
│ functional_2 (Functional)       │ ?                      │    18,191,008 │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ flatten_1 (Flatten)             │ ?                      │   0 (unbuilt) │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense_3 (Dense)                 │ ?                      │   0 (unbuilt) │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense_4 (Dense)                 │ ?                      │   0 (unbuilt) │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense_5 (Dense)                 │ ?                      │   0 (unbuilt) │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense_6 (Dense)                 │ ?                      │   0 (unbuilt) │
├─────────────────────────────────┼────────────────────────┼───────────────┤
│ dense_7 (Dense)                 │ ?                      │   0 (unbuilt) │
└─────────────────────────────────┴────────────────────────┴───────────────┘
 Total params: 18,191,008 (69.39 MB)
 Trainable params: 0 (0.00 B)
 Non-trainable params: 18,191,008 (69.39 MB)

Got Accuracy of 0.9575660824775696 on Test Dataset

Download My Fine Tuned InceptionV3 Model

• FineTuneInception.h5
• FineTuneInception.weight.h5
• To use them run

from tensorflow.keras import models

model = models.load_model("FineTuneInception.h5)

• If this repo helped in any way, give it a like
• Author : @shoryasethia