Error while processing rearrange-reduction pattern "(i j) c f h w -> c f (i h) (j w)". Input tensor shape: torch.Size([3, 16, 256, 256]). Additional info: {'i': 1}.

[Yaomingshuai]

[mumukawayi]

same problem

[wuleibegreat]

same problem

[mumukawayi]

I guess this may caused by unappropriate torch version. I met this problem and solved it by refering #15 to rebuild the environment.

[SSUHan]

Is this issue solved? I encounter the same problem. If we downgrade torch version from 2.0 to 1.13 <, xformers are not compatible with it :(

[mumukawayi]

Is this issue solved? I encounter the same problem. If we downgrade torch version from 2.0 to 1.13 <, xformers are not compatible with it :(

I solved it by following #15 to downgrade torch and rebuild the environment. xformers will conflict with the downgraded torch but anyway the code can still process and output the gif.

[SSUHan]

In my case, the model could output a gif but it is just blacked out like this: Is it OK in your case? @mumukawayi

[mumukawayi]

In my case, the model could output a gif but it is just blacked out like this: Is it OK in your case? @mumukawayi

In my case, the model can successfully produce the moving scene of the moonlight.

[SSUHan]

I finally find a way,

def _i(tensor, t, x):
    r"""Index tensor using t and format the output according to x.
    """
    shape = (x.size(0), ) + (1, ) * (x.ndim - 1)
    return tensor.to(x)[t].view(shape).to(x)

This works for me. But I still wonder why the below code does not work:

def _i(tensor, t, x):
    r"""Index tensor using t and format the output according to x.
    """
    shape = (x.size(0), ) + (1, ) * (x.ndim - 1)
    tensor = tensor.to(t)
    return tensor[t].view(shape).to(x)

[handsomeme001]

After I made modifications to the code, the problem was resolved

def video_tensor_to_gif(tensor, path, duration = 120, loop = 0, optimize = True):
    tensor = tensor.permute(1,2,3,0)
    images = tensor.unbind(dim = 0)
    images = [(image.numpy()*255).astype('uint8') for image in images]
    # imageio.mimwrite(path, images, fps=8)
    # return images
    imageio.mimsave(path, images, 'GIF', duration = duration)
    return path

This works for me.

[Yinhance]

I finally find a way,

def _i(tensor, t, x):
    r"""Index tensor using t and format the output according to x.
    """
    shape = (x.size(0), ) + (1, ) * (x.ndim - 1)
    return tensor.to(x)[t].view(shape).to(x)

This works for me. But I still wonder why the below code does not work:

def _i(tensor, t, x):
    r"""Index tensor using t and format the output according to x.
    """
    shape = (x.size(0), ) + (1, ) * (x.ndim - 1)
    tensor = tensor.to(t)
    return tensor[t].view(shape).to(x)

I met the same error, would you mind tell me where you change this function?