sauradip / MultitaskNet

Joint Semantic segmentation and Depth Estimation

MultitaskNet implementation in PyTorch

Prerequisites

• Linux
• Python 3.7.0
• CPU or NVIDIA GPU + CUDA CuDNN

Getting Started

Installation

• Install PyTorch 0.4.1.post2 and dependencies from http://pytorch.org
• Clone this repo:

git clone https://github.com/imvinod/MultitaskNet
cd MultitaskNet
pip install -r requirements.txt

Dataset preparation

sunrgbd dataset

• Download and untar the preprocessed sunrgbd dataset under /datasets/sunrgbd

FuseNet train/test

visdom visualization

• To view training errors and loss plots, set --display_id 1, run python -m visdom.server and click the URL http://localhost:8097
• Checkpoints are saved under ./checkpoints/sunrgbd/

train & test on sunrgbd

python train.py --dataroot datasets/sunrgbd --dataset sunrgbd --name sunrgbd

python test.py --dataroot datasets/sunrgbd --dataset sunrgbd --name sunrgbd --epoch 100

Results

• We use the training scheme defined in MultitaskNet
• Loss is weighted for SUNRGBD dataset
• Learning rate is set to 0.0001 for SUNRGBD dataset
• Results can be improved with a hyper-parameter search

More details coming up !