Hey everyone,
I am currently working on a project using YOLOv8 for detecting objects in low-light or nighttime conditions. While the model performs decently in daylight, the accuracy drops significantly in dim environments, especially when detecting smaller objects like pedestrians or animals.
I have already tried using image augmentation (brightness/contrast) during training.., and I am also experimenting with pre-processing using histogram equalization. Still, results are inconsistent.
A few questions:
Are there recommended techniques or datasets specifically for low-light object detection: ??
Would transfer learning from a model trained on infrared or night vision data help: ??
How effective is training with synthetic low-light data: ??
I am using YOLOv8n due to hardware limitations (Jetson Nano).., but open to suggestions on balancing performance and speed.
Any guidance from the community would be truly appreciated. If you have tackled similar use cases or have tips to enhance low-light performance, I would love to hear your experience! I have also gone through this thread https://community.ultralytics.com/t/yolov8-not-detecting-small-objects-well-salesforce-training-in-pune
but still need some more help.
Thanks in advance,
Daniel