- Task 1 - Track 1 – Single-domain training (Models must be trained exclusively on the published Cityscapes dataset) - Method: PixOOD YOLO (="Model Selection")
- Method info
- Samples list
- Per sample details
method: PixOOD YOLO (="Model Selection")2024-08-23
Authors: Tomáš Vojíř , Jan Šochman , and Jiří Matas
Affiliation: Czech Technical University in Prague
Description: If we are "playing the benchmark" and post-hoc selecting what method to run on what subparts, this method is combination of the PixOOD w/ ResNet-101 DeepLab v3 for semantic segmentation part of the benchmark and PixOOD w/ DeepLab Decoder for SMIYF and synobj part.
This benchmark then basically boils down to two disjoint tasks: 1) semantic segmentation, and 2) road anomaly detection. For each of these tasks, there is clear sota from the respective fields that is currently unbeatable by single approach.