Deep learning to identify apparels
- 0 Collaborators
The good old MNIST dataset is the Hello-World dataset for deep learning with computer vision tasks. Many researchers use it to benchmark their findings. But time has changed. It is time to move on to Fashion MNIST which is a strong replacement for the original MNIST dataset. It is comparatively new. This project includes my exploration of the dataset, coming up with deep learning models and more. ...learn more
Project status: Under Development
Intel Technologies
AI DevCloud / Xeon,
Intel Python
Overview / Usage
Benchmarking deep learning results on the original MNIST dataset is becoming a thing of past. Hence the need for a comparatively different yet simple Hello-World dataset for deep learning researchers - Fashion MNIST. It is important to keep up the pace and get accustomed to the basic datasets in the field. the project is more for research than deployment or other business problems. To know more about Fashion MNIST, readers are requested to go here: https://github.com/zalandoresearch/fashion-mnist
Methodology / Approach
- Deep learning to tackle the original image classification task
- Fine-tuning and softmax smoothing to improve predictive performance
- Clustering, dimensionality reduction to understand the data
Technologies Used
- Python
- fastai, scikit-learn, matplotlib, PyTorch, pandas, numpy