The second project for Udacity Deep Learning Nanodegree program. In this project, I explored state-of-the-art CNN models for classification, built a pipeline to process real-world, user-supplied dog images. Given an image of a dog, my model will identify an estimate of the canine’s breed. If supplied an image of a human, just for fun, my code will identify the resembling dog breed.
The training data for this project is located here. This dataset contains 133 different breeds of dogs and is already split into train, test, and validation sets. Place the training, testing, and validation datasets in the dogImages folder.
Python: 3.6
PyTorch: 0.4.0
GPU: NVIDIA GeForce GTX 970M
Windows: 8.1
With the pretrained Resnet50, I used transfer learning on new dog image data, the models has 75% accuracy on test data for dog breed with only 5 epochs of training (took less than 10 mins). The final user app used the pretrained VGG16 as a dog detector, then used the tuned Restnet50 as dog breed classifier, for humans, I used OpenCV's implementation of Haar feature-based cascade classifiers to detect human faces in images.
I tried the app with a photo of myself, it showed I look like a Bull terrier and Trump looks like Greyhound.
here is an example of a Bull terrier:
For Trump:
here is an example of a Greyhound:
Check the dog_app.ipynb for model with detail in code.
Check myphoto for the photos I tried with my app.