Home

Awesome

FCHarDNet

Aug-15-2021 Update

cd CatConv2d/
python setup.py install

(Please note that backward path for CatConv2d hasn't been implemented)

<br>

Fully Convolutional HarDNet for Segmentation in Pytorch

Architecture

<p align="center"> <img src="pic/fchardnet70_arch.png" width="540" title="FC-HarDNet-70 Architecture"> </p>

Results

<p align="center"> <img src="pic/fchardnet70_cityscapes.png" width="420" title="Cityscapes"> </p>
Method#Param <br>(M)GMACs /<br> GFLOPsCityscapes <br> mIoUfps on Titan-V <br>@1024x2048fps on 1080ti <br>@1024x2048
ICNet7.730.769.56348
SwiftNetRN-1811.810475.5-39.9
BiSeNet (1024x2048)13.411977.73627
BiSeNet (768x1536)13.466.874.772**54**
FC-HarDNet-704.135.476.07053
FC-HarDNet-70 V2 <br />(with CatConv2d)4.135.476.09963

DataLoaders implemented

Requirements

Usage

Setup config file

Please see the usage section in meetshah1995/pytorch-semseg

To train the model :

python train.py [-h] [--config [CONFIG]]

--config                Configuration file to use (default: hardnet.yml)

To validate the model :

usage: validate.py [-h] [--config [CONFIG]] [--model_path [MODEL_PATH]] [--save_image]
                       [--eval_flip] [--measure_time]

  --config              Config file to be used
  --model_path          Path to the saved model
  --eval_flip           Enable evaluation with flipped image | False by default
  --measure_time        Enable evaluation with time (fps) measurement | True by default
  --save_image          Enable writing result images to out_rgb (pred label blended images) and out_predID

Pretrained Weights

Prediction Samples

<p align="center"> <img src="pic/sample01.jpg" width="800"> </p> <p align="center"> <img src="pic/sample02.jpg" width="800"> </p> <p align="center"> <img src="pic/sample03.jpg" width="800"> </p>