carla-net

Title TO DO:

Introduction

Generating datasets from real life environment is a vey challenging task , as there are a lot of variables involved which not only makes the process difficult but also consumes a lot of time and effort. To mitigate this and to accelerate the process of generating larger amounts of datsets in a shorter span of time, we here use a simulator called 'Carla', which is an open-source simulator for autonomous driving research. To know more about this please visit their website Carla Simulator

![Alt text](https://github.com/MRSD2018/carla-net/tree/icnet-init/Results/carla-black-m.png ![enter image description here](https://github.com/MRSD2018/carla-net/tree/icnet-init/Results/carla-black-m.png)"Optional title")

Leveraging the power of Carla to generate real enivronment scenes , we generated about 50 Gb of high qualitydata,where the envioronment had different weather conditions.

Problem we are solving

We use the dataset generated from Carla to perform transfer learning on an pre-trained network learned on citsycapes dataset and then we validate the performance on data from real world.

Approach

IC-net which is now state-of-the art in real time semantic segmetation was used to train on the dataset generated from the simulator.

Installation

• Install deep learning framework -Pytorch
• Install Carla simulator

Dataset

We used Cityscape data for this project. You can download it by going to this link CityScape.

Steps to be followed

Results

We see that the network is able to learn and perfrom reasonably well in the real world data.

Finetuning imagenet on carla simulation renderings