search

xavysp / DexiNed

DexiNed: Dense EXtreme Inception Network for Edge Detection
MIT License
738 stars 157 forks source link

Training parameters on my own datasets #99

Open fxcdl opened 2 years ago

fxcdl commented 2 years ago

Hi, thank you for your code, I want to train on my own datasets, how do i set the "mean_pixel_values" and are there other parameters that need to be modified

xavysp commented 2 years ago

Hi, maybe you can start training DexiNed with the BIPED settings. I think it will not change much. To see the difference you can get the mean of your dataset and check later.

Cheers.

fxcdl commented 2 years ago

Thank you for your reply, Why the parameters "mean_pixel_values" on BIPED i calculate is different from the code.

image

xavysp commented 2 years ago

Thank you for your reply, Why the parameters "mean_pixel_values" on BIPED i calculate is different from the code.

image

Can I see the results?

fxcdl commented 2 years ago

Thank you for your reply, Why the parameters "mean_pixel_values" on BIPED i calculate is different from the code. image

Can I see the results?

Thank you for your reply, This is my results: image

xavysp commented 2 years ago

Hi, hope you are enjoying your holiday :), well the results of my rgb_mean are from the BIPED image but before post-processing them (white balance, gamma correction,...). In other words, the raw biped images were considered for the mean.

Any way, I think you can use this new mean, let me know how they are working.

Cheers,

Xavier