This repository contains the code used in paper Robust Object Detection in Challenging Weather Conditions to generate weather effects synthetically using
1. Analytical Methods
2. Cyclic GAN
3. Neural Style Transfer
We generate the foggy, rainy and snowy weather using the following pipeline. To change the intensity of the weather, parameters like gaussian noise, blurring etc could be changed in the Juyter notebooks Fog/Rain/Snow_Effect_Generation.ipynb.
We trained three VGG classifier for fog-clear, rain-clear and snow-clear weather classification. The pretrained weights are available here.
We trained three Cyclic-GAN for clear<->fog, clear<->rain and clear<->snow I2I translation. The pretrained weights are available here.