- Download dataset
- Investigate dataset
- Preprocess images
- Split dataset on train/validation parts
- Train simple classificator and evaluate it
- Know what neural networks means
- Create small fully connected NN and train it
- Compare results with simple classificator
- Find what convolution layers means
- Study some optimization techniques - dropout, batchnorm, etc
- Explore varios data augmentation methods
- Build deep NN
- Is it works better?
- What about size/speed?
- Choose one of the size reduction method and implement it:
- pruning, quantization
- knowledge distillation
- XNOR-net
- you own method?
- Visualizing what ConvNets learn(optional)
- Compare with all other teams