# ML Assignment 2 Andrey Chertkov ## Transform images to same size Source image ![](https://i.imgur.com/IZ9c2e8.png) After padding ![](https://i.imgur.com/hEHf6GT.png) ## Frequency Before augmentation ![](https://i.imgur.com/EW8TYWE.png) After ![](https://i.imgur.com/iRCOCIJ.png) ## Augmentation I choose rotation and random noise, because other methods has a cons. * Fliping will not work for sign of turns * Crop will not work because size of image is small. * Changing hue not work because color of sign is matter Before ![](https://i.imgur.com/3bc1Lq2.png) After ![](https://i.imgur.com/07eJJvg.png) ## Evaluatoin Validation ![](https://i.imgur.com/7OyTPAF.png) Test ![](https://i.imgur.com/6Mqx9Rc.png) Precision and recall ![](https://i.imgur.com/v1i57N0.png) Accuracy on not augmentat data 0.65 ## Plots ![](https://i.imgur.com/VtAG1y0.png) ![](https://i.imgur.com/3Gdt07K.png) ![](https://i.imgur.com/FIJyRyx.png) Where size is a [Size X Size]