dataset
probing_visual_concepts_in_traffic_scenes
D2ICE-Automotive-Research
or hover any field below to flag it
Overview
Name
probing_visual_concepts_in_traffic_scenes
Source
D2ICE-Automotive-Research
Episodes
0
Robot count
0
Format
other
Description
Probing toolkit for lightweight vision-language models in automated driving: extract layerwise VLM features from traffic-scene images, train linear probes to analyze concept encoding (presence/count/spatial/orientation), and benchmark direct VQA performance on a CARLA-based counterfactual dataset.
Robots used
null
Links
HuggingFace dataset
null