This page contains ideas on the design of fitness-guided metrics in V2XGen's workflow. Here we provide the insights for the Fop and Flp.
Fitness Metric
The core idea of this guidance strategy is to generate scenes with occlusion relationships and long-range objects in the perspective of the ego vehicle while maximally confusing cooperative perception systems. Here are the insights:
A cooperative perception system should help the ego vehicle detect objects that interfere with the ego vehicle's line of sight. We hope to find as many scenarios as possible where the cooperative perception system does not help the ego vehicle successfully detect occluded objects.
A cooperative perception system should help the ego vehicle detect objects far from the ego vehicle. We hope to find as many scenarios as possible where the cooperative perception system does not help the ego vehicle successfully detect distant objects.
Based on the insight 1, we designed Fop. Please refer to our research paper for a detailed explanation of the symbols.
Fop tries to find objects with low occlusion rates in the CV perspective and high occlusion rates in the ego perspective, but the cooperative perception system ultimately fails to detect objects.
Based on the insight 2, we designed Flp. Please refer to our research paper for a detailed explanation of the symbols.
Flp tries to find objects that are close to the CVs and far away from the ego vehicle, but the cooperative perception system ultimately fails to detect objects.