Computer Vision final project for COMPSCI 670 done by Shubham Mukherjee and Deep Chakraborty.
In this project, we experiment with semantic segmentation using Dilated Conv Nets [1] on the Stanford Background Dataset [2]. Key contributions in this project:
- Code for end-to-end Training of Front end + Context module
- Training using Batch Norm [3] in the Context module to enable random initialization and eliminate the need for careful initialization techniques
- Fine-Tuning the PASCAL VOC pre-trained models on the Stanford Background Dataset to obtain near state-of-the-art accuracy.
This repository is built upon : https://github.com/nicolov/segmentation_keras
@TODO : To be modified as a Fork of segmentation_keras repo
Pre-trained model:
curl -L https://github.com/nicolov/segmentation_keras/releases/download/model/nicolov_segmentation_model.tar.gz | tar xvf -
Install dependencies:
pip install -r requirements.txt
# For GPU support
pip install tensorflow-gpu==1.3.0
Prediction:
python predict.py --weights_path conversion/converted/dilation8_pascal_voc.npy
Download Stanford Background Dataset from:
http://dags.stanford.edu/data/iccv09Data.tar.gz
We have to pre-process the Stanford dataset...