search

dlam4h / OCD-PyTorch

The code for paper 《Object Contour Detection with a Fully Convolutional Encoder-Decoder Network》
19 stars 4 forks source link

nothing else,thanks for your code #2

Open JunlinHan opened 4 years ago

JunlinHan commented 4 years ago

,

harshitsankhla commented 4 years ago

@JunlinHan If possible, can you please share the dataset you used for training ? Also which code did you try, 1 or 2 ?

JunlinHan commented 4 years ago

@JunlinHan If possible, can you please share the dataset you used for training ? Also which code did you try, 1 or 2 ?

Hi, I used both PASCAL VOC and BSDS500. From my perspective, the second one performs better.

harshitsankhla commented 4 years ago

Do you train with the dataset images as they are or do some pre-processing. I basically wanted to know if I should black out all the non-contour pixels or just leave them as they are in the segmentation mask.

JunlinHan commented 4 years ago

Do you train with the dataset images as they are or do some pre-processing. I basically wanted to know if I should black out all the non-contour pixels or just leave them as they are in the segmentation mask.

You should leave them. They are groundtruth hence there's no need to change them.

harshitsankhla commented 4 years ago

I see. Thanks for the information.