/ZeroshotSemanticSegmentation

Zeroshot learning for semantic segmentation using joint visual-semantic embedding space.

Primary LanguagePython

Zero-Shot Semantic Segmentation

  • Princeton University Senior Thesis
  • Advisor: Professor Olga Russakovsky, Princeton University Department of Computer Science
  • Part of the Princeton VisualAI Lab

Project Overview

This is the PyTorch implementation of the seenmask zeroshot network (SZN) described in Rohan Doshi's senior thesis "Zero-shot Semantic Segmentation." Please reference this paper (rohan_doshi_senior_thesis.pdf) to understand the code.

Installation

** Requirements: ** Conda (with Python 3) and Linux

  1. Install Conda

  2. Clone repository

git clone https://github.com/RohanDoshi2018/ZeroshotSemanticSegmentation.git
cd ZeroshotSemanticSegmentation
  1. Create new conda environment
conda create --name thesis_env
  1. Install Dependencies
conda install pytorch torchvision -c pytorch
pip install pytz pyyaml scipy fcn jupyter tensorflow tensorboardX

  1. Activate your conda environment source activate thesis_env

  2. Run code

./train.py -c 4 -g 0
  1. [Optional] Run Tensorboard Server. Use Ngrok tunnel to access server remotely.
tensorboard --logdir /opt/visualai/rkdoshi/ZeroshotSemanticSegmentation/tb