Used pretrained CNNs and custom CNNs to classify dog images in project 1 and dog breeds in project 2. Transfer learning is a popular method in computer vision because it allows us to build accurate models in a timesaving way. These projects were in my Udacity's Nanodegree courses.
Notebook exploring torchvision pretrained models for image classification. View here
CNNs used :
- VGG16
- AlexNet
- ResNet
Run following scripts:
- First change the directory and add image in the uploaded_images folder if you want,
- Run the following script
./run_models_batch_uploaded.sh
If the permission denied error shows up run the following before step 2
chmod u+r+x run_models_batch_uploaded.sh
In this prohect objective is to classify dog breeds using pretrained network such as ResNet and custom network architecture for comparable study. Also a exploration using HarCascade from OpenCV for face detection is performed.
- Python 3.6+
- Pytorch (version 0.4+)
- OpenCV
- Numpy
- Increasing Accuracy
- Using more networks
Udacity Deep Learning Nano Degree