Open CuddleSabe opened 2 weeks ago
Hi, Our approach has limitations under extreme low-light or backlit conditions, due to the gamma correction-driven design of CPGA-Net (both pretrained branches). Under these specific conditions, it typically produces warm colors, which correspond to the color shift you mentioned, as observed in the white regions of the figures you provided. Moreover, we tested our approach on benchmarks and driving scenes, which did not include human-related scenarios. As a result, our approach may not perform as well in these contexts.
Further adjustments are needed to better handle challenging lighting conditions.
Thanks for sharing your feedback!
Hi, I use the pretrained weight on real data. the results have very serious color shift