How to create dataset like LOLv2 real or Syn for finetunning

Fediory / HVI-CIDNet

"You Only Need One Color Space: An Efficient Network for Low-light Image Enhancement"

MIT License

190 stars 16 forks source link

How to create dataset like LOLv2 real or Syn for finetunning #29

Open DevilANANDGupta opened 3 months ago

DevilANANDGupta commented 3 months ago

Hi @Fediory I have some images captured from smartphones that are well-lit, and I want to convert these images into low-light versions for fine-tuning HVI-CIDNet. I tried adjusting the brightness using cv2.convertScaleAbs, which helped modify the brightness, but after training for 1000 epochs, I achieved an SSIM of 0.9465 and a PSNR of 17.625. However, when I performed inference on unseen images, the output had significantly higher luminance than expected.

Fediory commented 3 months ago

Please give more details. What your train datasets? is it paired or not? What is your test images? Would you like to share some of the examples to me?