Implementation of several Semantic Segmentation architectures for common segmentation tasks.

This respository consists of the code for Semantic segmantation using several Deep Learning Architectures.

• If space complexity and time complexity is taken into criteria, ENET is a viable option. It can be deployed even in mobile devices.
• If the time and space complexity can be relaxed, UNET and Modified VGG net performs better than their other variant ENET.
• The training code is very much scalable towards any new architecture.
• All changes made in the config file will effect in the training process so that the training logic can be without hassle.
• The training configuartion are easily tunable through the config file provided.

• The training module has been built using Pycharm 2018.1.4.
• The System requirement’s are 2.7 GHz Intel Core i5 with atleast 8 GB of RAM.

You can use Anaconda to install opencv with the following command line.:

conda install -c conda-forge opencv

You can use PIP to install the module imgaug with the following command line.:

pip install imgaug

You can use PIP to install tensorflow with the following command line or please go through their official installation guideline

pip install tensorflow

You can use PIP to install keras with the following command line or please go through their official installation guideline

pip install keras

Run the following script to dispatch the trainer.

python3 train.py  --conf=./config.json

Don't feel shy to drop a star, if you find this repo useful. I would love for you to contribute to Semantic Segmentation, check the LICENSE file for more info.

Stanly Moses – @Linkedin – [email protected]

Distributed under the MIT license. See LICENSE for more information.

https://github.com/StanlyHardy/Semantic-Segmentation