seoungwugoh / opn-demo Goto Github PK

I am facing this Error while running inference on custom images,
it'll be much appreciated if you help me solve the error,
Thank you

Hi author,
It would be very helpful if the training code can be released, we'd like to train it. Thanks.

Hi, can you be more specific about how to divide videos into clips? Some deep neural network?

Hi, I was wondering whether you will release the code of training part? Thanks.

Hi, I follow your papers "Onion-Peel Networks for Deep Video Completion" and "A Unified Model for Semi-supervised and Interactive Video Object Segmentation using Space-time Memory Networks". The papers all describe the encoder which output key and value, but don't konw how to propel encoder to output key(for addressing) and value (stores detailed information), what is the loss function? how to train encoder?
I appreciate it if you reply it .
thank you！

tim.credy in China

1.

Is each frame - mask pair bound together during the whole training? Or they may have different mate at each iteration?

The later way may be an augmentation of data. But I'm afraid that with this augmentation, the model would learn the inpainting from already-seen samples before (of the same frame but different mask), instead of really inferring it.

2.

When and where is the total variation loss (L_tv in the paper) computed? Before merging the output peel pixels with the non-hole region or after? Only on the peel or on the whole image?

According to your test strategy, the calculation graph will be huge during your training. How do you solve this problem?

Hi, great work on video Inpainting!
I have two questions regarding the training of the network.

1. When training the network for video inpainting, how many reference images are needed, and how did you sample the frames?
2. Do you optimize the network for every peel or sum up all the losses and optimize after the hole region is completed?

Thank you.

@seoungwugoh Thanks for releasing the codebase.
How did you train the TCN module? Could you explain in detail about training strategy and losses?

RuntimeError: Subtraction, the - operator, with a bool tensor is not supported. If you are trying to invert a mask, use the ~ or logical_not() operator instead.