Cars-or-Humans-Classifier

Classification model that uses a simple convolutional neural network to distinguish humans from vehicles. Made using Tensorflow and Keras. The model is trained with 9.532 images depicting humans and 8.144 images depicting cars.

Human dataset: (http://vision.stanford.edu/Datasets/Stanford40_JPEGImages.zip)

Cars dataset: (http://ai.stanford.edu/~jkrause/car196/cars_train.tgz)

Model architecture

Model: "sequential"
_________________________________________________________________
 Layer (type)                Output Shape              Param #
=================================================================
 conv2d (Conv2D)             (None, 298, 298, 16)      448

 max_pooling2d (MaxPooling2D  (None, 149, 149, 16)     0
 )

 conv2d_1 (Conv2D)           (None, 147, 147, 32)      4640

 max_pooling2d_1 (MaxPooling  (None, 73, 73, 32)       0
 2D)

 conv2d_2 (Conv2D)           (None, 71, 71, 64)        18496

 max_pooling2d_2 (MaxPooling  (None, 35, 35, 64)       0
 2D)

 conv2d_3 (Conv2D)           (None, 33, 33, 64)        36928

 max_pooling2d_3 (MaxPooling  (None, 16, 16, 64)       0
 2D)

 flatten (Flatten)           (None, 16384)             0

 dense (Dense)               (None, 512)               8389120
 dense_1 (Dense)             (None, 1)                 513

=================================================================
Total params: 8,450,145
Trainable params: 8,450,145
Non-trainable params: 0
_________________________________________________________________

Demonstration with random images from Google

Image of a car

Output:

1/1 [==============================] - 0s 250ms/step
This is a car

Image of a human

Output:

1/1 [==============================] - 0s 141ms/step
This is a human

Let's see if a picture of a person whose legs are concealed will generate the correct output;

Output:

1/1 [==============================] - 0s 139ms/step
This is a human

What if there are both cars and humans in a picture?

Output:

1/1 [==============================] - 0s 139ms/step
This is a human

Since this is not a multi label classifier the most prominent object in this picture is recognized which in this case is the human.

Also, the model is for now unable to 'understand' images that don't contain humans or cars, thus results are unexpected. Feel free to ontact me if you have any questions or suggestions.