Algorithm Details

WD2 Panoptic

seamseg_mvd_ps

Baseline Panoptic submission Mapillary Vistas

Seamless Scene Segmentation Panoptic Segmentation trained with altered MVD to include new WildDash 2 labels. Trained from a ResNet50 pretrained on ImageNet as a WildDash 2 baseline by Oliver Zendel/AIT

Mapillary Vistas dataset with added van/pickup truck relabeling

300 epochs; max. image size FullHD, batch size 2

python/pytorch

Lambda Labs Hyperblade Titan RTX

Source code

Seamless Scene Segmentation

Porzi, Lorenzo and Rota Bulò, Samuel and Colovic, Aleksander and Kontschieder, Peter

CVPR 2019

@InProceedings{Porzi_2019_CVPR,
author = {Porzi, Lorenzo and Rota Bul\`o, Samuel and Colovic, Aleksander and Kontschieder, Peter},
title = {Seamless Scene Segmentation},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2019}
}