Adapting CNNs for Fisheye Cameras without Retraining

We introduce a pre-processing and fine-tunning free approach to adapting existing pre-trained networks to fisheye imagery. We present this in our paper:

Paper: Adapting CNNs for Fisheye Cameras without Retraining

Authors: Ryan Griffiths, Donald G. Dansereau

Project Page: roboticimaging.org/Projects/RectConv/

Setup

Environment

Get code and build docker image (requires docker to be installed):

git clone https://github.com/RoboticImaging/RectConv.git
cd RectifyConv
docker build -t rect_conv .

Dataset

Download the woodscape dataset here

Checkpoints

Download checkpoints of pre-trained networks:

• Deeplabv3_Resnet50
• Deeplabv3plus_Resnet101
• FCN_Resnet50
• FCN_Resnet101

Run

Run the code inside container:

docker run --rm \
-v .:/workspace \
--shm-size 4G \
--gpus all \
rect_conv \
python run.py --model deeplabv3plus_resnet101 --data_path datasets/woodscape --model_checkpoints trained_models/deeplabv3plus_resnet101_cityscape.pth

Note: to run the command above the dataset should be stored in a ./datasets directory and the checkpoints should be stored in a ./trained_models directory

A .devcontainer folder is also provided if vscode used.

Citation

If you find our work useful, please cite the below paper:

@article{griffiths2024adapting,
  title = {Adapting CNNs for Fisheye Cameras without Retraining},
  author = {Ryan Griffiths and Donald G. Dansereau},
  journal = {arXiv preprint arXiv:2404.08187},
  URL = {https://arxiv.org/abs/2404.08187},
  year = {2024},
  month = apr
}