View article

In this technical report, we briefly describe the solutions of our “gxj” team in CVPR 2024 PBDL: Raw Image Based Over-Exposure Correction Challenge. In this task, we propose an efficient framework for correction of overexposure based on original images. Specifically, first to Raw image data preprocessing, in order not to reduce the performance of the model itself, the image of different exposure ratio conversion into RGB format for training, and then based on the effective area perception exposure correction network, through adaptive learning and bridging different area exposure representation to process mixed exposure. Subsequently, we employed an effective image super-resolution model on the corrected resulting maps at 2x super-resolution. Our framework is able to handle different types of overexposed dataset provided during the final testing phase and rank first on the final leaderboard.