Batch inference error

[JCBrouwer]

Hello,

Thanks for this port, the results are great so far!

One issue I've run into is that running batches does not work. Running the test script from the README with a batch size of more than one leads to the error below:

import torch

device = torch.device('cuda')
precision = torch.float16

model = torch.jit.load('film_net_fp16.pt', map_location='cpu')
model.eval().to(device=device, dtype=precision)

img1 = torch.rand(2, 3, 720, 1080).to(precision).to(device)
img3 = torch.rand(2, 3, 720, 1080).to(precision).to(device)
dt = img1.new_full((2, 1), .5)

with torch.no_grad():
    img2 = model(img1, img3, dt)

RuntimeError: The following operation failed in the TorchScript interpreter.
Traceback of TorchScript, serialized code (most recent call last):
  File "code/__torch__/util.py", line 21, in forward
  for _5 in range(torch.len(pyramid)):
    image = pyramid[_5]
    _6 = torch.append(_4, torch.mul(image, scalar))
                          ~~~~~~~~~ <--- HERE
  return _4
def concatenate_pyramids(pyramid1: List[Tensor],

Traceback of TorchScript, original code (most recent call last):
  File "C:\Users\Danylo\PycharmProjects\frame-interpolation-pytorch\util.py", line 105, in forward
    # multiplied with a batch of scalars, then we transpose back to the standard
    # BxHxWxC form.
    return [image * scalar for image in pyramid]
            ~~~~~~~~~~~~~~ <--- HERE
RuntimeError: The size of tensor a (512) must match the size of tensor b (2) at non-singleton dimension 3

The easy fix is to just split and run individually, but I assume batch inference would be a bit faster (not that this model seems to be particularly heavy).

[dajes]

Hi @JCBrouwer. With the release 1.0.2 it is now fixed