[Paper]
Our paper, Cloud Object Detector Adaptation by Integrating Different Source Knowledge (COIN), has been accepted by NeurIPS 2024.
We follow the trend of the times and explore an interesting and promising problem, Cloud Object Detector Adaptation (CODA), where the target domain leverages detections provided by a large vision-language cloud detector to build a target detector.
Please note that CODA does not restrict whether CLIP is used, even though CLIP is used in our method COIN.
Code will be available soon!