Unable to run train_online.py about osvos-pytorch

[scaelles]

We use volatile=True to use less memory at test time as no gradient has to be computed. The message you get is just a warning so it should work anyway. I recommend you to use the latest release (0.3.1) so you don't have to worry about changes in the API done in the version that you use now.

[mrteera]

I'm not sure if I fix it correctly:

https://github.com/kmaninis/OSVOS-PyTorch/blob/master/layers/osvos_layers.py#L55

    return F.pad(x, [
        crop_w.ceil().int()[0], crop_w.floor().int()[0],
        crop_h.ceil().int()[0], crop_h.floor().int()[0],
    ])

to

    return F.pad(x, (int(crop_w.ceil()[0]), int(crop_w.floor()[0]), int(crop_h.ceil()[0]), int(crop_h.floor()[0])))

[scaelles]

Hello mrteera,
The fix that you have applied looks good to me. We have only tested our code in stable released versions of Pytorch (0.2, 0.3, 0.3.1) and we haven't encountered that error previously so I guess they have change the behavior of the pad function. If the code works after your fix, please close the issue.
Best

[mrteera]

I have another error. I'm not sure if it comes from my fix or not.

[Epoch: 10000, numImages:     1]
Loss: 1109.525879
Online training time: 1661.821477145073
Testing Network
train_online.py:182: UserWarning: volatile was removed and now has no effect. Use `with torch.no_grad():` instead.
  inputs, gts = Variable(img, volatile=True), Variable(gt, volatile=True)

[kmaninis]

Hi, once PyTorch 0.4 is stable and released, we will make all our code compatible to that. I hope this helps.