How to train on a custom dataset about graphonomy HOT 5 CLOSED

Hi,
I think maybe you can just use Deeplab to do it. You can refer to the Deeplab which may be good enough on your custom dataset.

from graphonomy.

Thanks, I just need to know how can I include watches and belts as its classifying it as background.
Does this segment human before classifying it further to other classes?

from graphonomy.

Hi，

1. You need to train/fine-tune the model with the label of watches and belts.
2. Our model does not segment human. You can use Deeplab or add a branch to segment human. It is also needed to train/fine-tune the model. Maybe you can try to use the results of the person-part segment to segment human.

from graphonomy.

Ok, I found one dataset for my needs
MHP v2
So i have formatted the data like CIHP

1. it has 59 classes so would it be a problem if I train it on transfer learning and how to correctly load it
2. I am getting 1 channel error but my images are 3 channel rgb
   ValueError: Expected more than 1 value per channel when training, got input size torch.Size([1, 256, 1, 1])
   Thanks

from graphonomy.

Hi,

1. For quickly train on it, you can change the `n_classes' to 59 in the here.
2. Can you give me more information about the ValueError? Thanks.

from graphonomy.