Transfer learning Project
- Implemented transfer learning with state of the art neural network architectures like ResNet50, InceptionV3, VGG16 to identify and categorize images of 10 different animal species.
- Image Augmentation is done to reduce overfitting and improve overall model validation accuracy
- Technologies used: Python, Keras, tensorflow
I have used the Animals-10 dataset publicly available on kaggle. The dataset contains 10 different classes of animals(dog, cat, horse, spyder, butterfly, chicken, sheep, cow, squirrel, elephant.). Each different class has 2000 - 5000 pictures. The total number of pictures in that dataset is around 28,000. This photos were taken from google.
| Model | Training Data | Accuracy |
|---|---|---|
| vgg16 | Img1000 | 82.92% |
| ResNet50 | Img1000 | 92.8% |
| ResNet50 + Fine Tuning | Img5000 | 96.09% |
| InceptionV3 | Img1000 | 96.66% |
| InceptionV3 + FineTuning | Img5000 | 97.81% |
This repository only contains the test results and Codes of the separate models. For detailed files and models please visit https://drive.google.com/drive/folders/13Cl3aQ5qnSDtmp0uMGLr7yK96_j2DvEI?usp=sharing