DeepLearning Project: Cloud Formation Classification

Data

The data was gathered from this Kaggle compatition Understanding Clouds from Satellite Images

Prerequisites

• Python 3.6
• TensorFlow V2.x (For Perceptron/SimpleAnn (used in main.py) TF V1.15)
• Opencv 4.x
• Pandas
• matplotlib
• dataclasses
• tqdm

Contents

• Prepare Data
• Single/Multi-layer NN
• CNN
• AutoEncoder/KNN
• Auxiliary Loss

Prepare Data

First dowload the data from Understanding Clouds from Satellite Images and unpack it to the data folder. Then run

python data/data_gen.py

This will extract the data into folders by there class, where each image contains one cloud formation only.

Single/Multi-layer NN

For now, main.py works with TF 1.15

To use run the SLP or MLP, run the main.py with the following arguments:

usage: 
python main.py [-h] --model MODEL [SLP,ANN,CNN] [--batch_size MINI_BATCH]
               [--samples SAMPLES] [--use_gpu GPU] 
               [--gpu_full FULL_GPU] [--weights WEIGHTS_PATH]

CNN

To use run the CNN, run the CNN.py:

usage:
python CNN.py

AutoEncoder/KNN

To use the AutoEndocer/KNN, run:

python autoEncoder.py
python classify_knn.py --model [PATH_TO_SAVED_MODEL]/encoder --images PATH_TO_MINI_DATA

Auxiliary Loss

To use the final (best results) model with the AE and auxiliary loss run:

python auxiliary_loss.py

Authors

Naomi Tal Tsabari
Shai Aharon