Kaggle-Dogs-vs-Cats

Writing an algorithm to classify whether images contain either a dog or a cat.

Requirements:

• python3
• keras
• tensorflow
• numpy V1.19.2
• scikit-image V0.17.2
• opencv-python
• matplotlib V3.2.2
• imutils

CNN architecture implemented in this repository:

3. AlexNet network structure table summary:

Layer Type	Output Size	Filter Size / Stride
INPUT IMAGE	227x227x3
CONV	55x55x96	11x11/4x4, K=96
ACT	55x55x96
BN	55x55x96
POOL	27X27X96	3x3/2x2
DROPOUT	27x27x96
CONV	27x27x256	5x5, K=256
ACT	27x27x256
BN	27x27x256
POOL	13X13X256	3x3/2x2
DROPOUT	13x13x256
CONV	13x13x384	3x3, K=384
ACT	13x13x384
BN	13x13x384
CONV	13x13x384	3x3, K=384
ACT	13x13x384
BN	13x13x384
CONV	13x13x256	3x3, K=256
ACT	13x13x256
BN	13x13x256
POOL	6X6X256	3x3/2x2
DROPOUT	6x6x256
FC	4096
ACT	4096
BN	4096
DROPOUT	4096
FC	4096
ACT	4096
BN	4096
DROPOUT	4096
FC	1000
SOFTMAX	1000

Kaggle Dogs vs Cats Dataset:

The dataset contains 25,000 images of dogs and cats

Evaluations of the Trained Networks:

• Training AlexNet from scratch: 90% Accuracy

• Fine tuning on ResNet

References:

• Deep Learning for Computer Vision with Python VOL1 & VOL2 by Dr.Adrian Rosebrock