pfnet-research / chainer-deepfill Goto Github PK

@msakai @delta2323 @iwiwi @okapies @mitmul
Hi,

Awesome Work!
I have some question while training the model,an error occurred as follow:

My chainer version is 5.3.0

In addition,I wonder if we could use the irregular mask put forward by partial convolution as input rather than use the free form mask.How could we solve it?

Last but not the least,I have noticed that the EDGE_FLIST is none,I wonder you say there are no influences whether EDGE_FLIST exist or not.How to get the EDGE_FLIST?

Looking forwarding to receiving your reply!I would very appreciate if you could help me!

Thanks,

Wondeful work.I have trained the model using your code,but I wonder how to print the mask and incomplete image during one batch.In test.py,I have set batchsize to 1.And here is my code:

out = batch_complete.data.get()
out = batch_postprocess_images(out, 1, 1)
Image.fromarray(out).save(args.eval_folder + "/Iout.png")
#failed
# mask = 1. - mask[:, :1]
# mask = mask.get()
# mask = batch_postprocess_images(mask, 1, 1)
# Image.fromarray(mask).save(args.eval_folder + "/mask.png")
gt = batch_pos.get()
gt = batch_postprocess_images(gt, 1, 1)
Image.fromarray(gt).save(args.eval_folder + "/Igt.png")
incomplete = batch_incomplete.get()
incomplete = batch_postprocess_images(incomplete , 1, 1)
Image.fromarray(incomplete).save(args.eval_folder + "/incomplete.png")
edge = batch_incomplete - mask[:, 1:] + mask[:, :1]
edge = edge.get()
edge = batch_postprocess_images(edge, 1, 1)
Image.fromarray(edge).save(args.eval_folder + "/edge.png")
#failed
# mask =  mask[:,:1]
# mask = mask.get()
# mask = batch_postprocess_images(mask, 1, 1)
# Image.fromarray(mask).save(args.eval_folder + "/mask.png")
x1 = x1.data.get()
x1= batch_postprocess_images(x1, 1, 1)
Image.fromarray(x1).save(args.eval_folder + "/x1.png")

I guess it may be related to the channel,but how to solve it?

The incomplete picture I get is covered by gray region rather than white region,how to replace the unknown area with white feature?

Looking forward to receiving your reply,thank you.

Hi,

Thank you very much for sharing the codes. I have one question:
If I train the v2 model with edge input, can I test the model without edge input? Just like the normal image inpainting system without guided user input.

Thank you very much!

Thank you for this incredible implementation of DeepFill.

Apologies for the stupid question, but I have been unable to implement the ability to restore from checkpoints when training.

MODEL_RESTORE seems to imply this exists out of the box, but I haven't been able to work it out.

Any help would be really appreciated.

Again, this is an amazing implementation of Deepfill.

I wonder if it is possible to use custom masks for evaluation?

For example: Use a png or jpeg mask in the command line to remove a person or object from a nature photo.

I'm happy to make a donation if you're able to add this function :-)