🚘 The easiest implementation of fully convolutional networks

• Task: semantic segmentation, it's a very important task for automated driving

• The model is based on CVPR '15 best paper honorable mentioned Fully Convolutional Networks for Semantic Segmentation

Results

Trials

Training Procedures

Performance

I train with two popular benchmark dataset: CamVid and Cityscapes

dataset	n_class	pixel accuracy
Cityscapes	20	96%
CamVid	32	93%

Training

Environment

# Install python3
conda create -n fcn python=3.9

# To activate this environment
conda activate fcn

# Install dep
conda install pytorch torchvision torchaudio pytorch-cuda=11.8 -c pytorch -c nvidia

and download pytorch 0.2.0 from pytorch.org

and download CamVid dataset (recommended) or Cityscapes dataset

CamVid 下载

https://tianchi.aliyun.com/dataset/128347?t=1720103543115

Cityscapes 下载

https://tianchi.aliyun.com/dataset/91542?spm=a2c22.28136470.0.0.448e4a0azc0Kkg&from=search-list

Run the code

• default dataset is CamVid

create a directory named "CamVid", and put data into it, then run python codes:

python3 python/CamVid_utils.py 
python3 python/train.py CamVid

• or train with CityScapes

create a directory named "CityScapes", and put data into it, then run python codes:

python3 python/CityScapes_utils.py 
python3 python/train.py CityScapes

Author

Po-Chih Huang / @pochih