In this project i implemented fcn32s and fcn8s from scratch for the task of semantic segmentation . i trained the model on CamVid dataset.
result metrics - iou in fcn 8s - 0.3872870561793166 mean accuracy in fcn 8s - 0.7741428788965246 pixelwise accuracy in fcn 8s - 0.8902169921874998
iou in fcn 32s - 0.33139500959453305 mean accuracy in fcn 32s - 0.7245959772991014 pixelwise accuracy in fcn 32s - 0.8706499348958332
The models are available through the kaggle notebook . kaggle notebook link - https://www.kaggle.com/code/shivamguptakaggle/fcn-segmentation-1-final
original paper - https://arxiv.org/abs/1411.4038