Datasets and Code will be released after acceptance.
The repo is the official implementation of "Detail-Preserving and Diverse Image Translation for Adverse Visual Object Detection".
This work aims to improve the performance of object detection models under complex lighting and adverse weather conditions by training them with detailed and diverse synthetic images generated by GANs. State-of-the-art results have been achieved on two public datasets, VisDrone and BDD100K.
Illustration of our task and approach.
Qualitative results of PAIT and baseline on BDD100K.