Week 1 — Warmth of Image
Title: Weather Condition Prediction from Image
Team Members: Berk GÜLAY, Samet KALKAN, Mert SÜRÜCÜOĞLU
E-mails Respectively: berkgulay.cs@gmail.com , abdulsametkalkan@gmail.com , mertsurucuogluu@gmail.com
Brief Explanation:
In our generation ML is very widely used area to predict very useful things and inference something from data(image etc.). In our project we aim to predict weather condition in given image by using various image descriptors and features. We hope to cover a large label range to classify images.
Sample Classes:
- Rainy
- Sunny
- Cloudy
- Foggy
- Snowy
- Thunderstorm / Light Thunderstorm / Lightning
DataSets:
Also some webcam images(not labeled and should be labeled) and photoshopped(as we want, so labeled) images can be used for data variation.
If you have suggestions or some useful resources/sites you would always be welcome to contact with us by given e-mails above. Thanks in advance.
Approach / Brief Algorithm Overview:
Various image descriptors(Hog,Color Histogram etc.) , features and frameworks(Caffe etc.) will be used to describe&classify images. Different Neural Net. architectures, SVM algorithms and Decision Trees will be tried for classification. Crop/eliminate images to make them feasible&same sized, normalization, regularization and these kind of basic applications will be applied and also new ones will be researched for true labeling performance. If it is possible to collect temperature/humidity/time features for every image in dataset, we will try to obtain them as basic features about images and use them in our classification algorithm as well. Best algorithm to predict correct labels will be used as latest model and it will be introduced as result. Stand by us!
