NVlabs / few-shot-vid2vid

Pytorch implementation for few-shot photorealistic video-to-video translation.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Problems with optical flow

zhhezhhe opened this issue · comments

I checked the optical flow between two neighboring frames in a video, and the result is:

[tensor([[[[[ 10.2950,   0.7181,  -4.7409,  ...,   3.1726,   6.1062,
             -8.1125],
           [ -3.6843,  -6.1567,  -7.4636,  ...,   3.6077,   2.9878,
              5.3008],
           [ -7.2534,  -6.3700,  -5.7260,  ...,   6.1592,   4.6107,
              6.7087],
           ...,
           [-13.3908, -15.8770, -17.8833,  ...,   9.0593,   9.0684,
              7.5918],
           [-14.2425, -15.4943, -16.1309,  ...,   8.8715,   7.9123,
              5.1580],
           [-15.3805, -18.3000, -18.8595,  ...,   9.3637,   3.7064,
              7.0863]],

          [[-31.9163, -33.5100, -46.0788,  ..., -57.1201, -48.2610,
            -35.4169],
           [-31.5390, -48.7966, -54.5237,  ..., -60.5762, -55.1705,
            -48.9994],
           [-46.4077, -52.8530, -54.1906,  ..., -56.5632, -57.1629,
            -55.7577],
           ...,
           [-29.6735, -33.5224, -34.8830,  ..., -34.7638, -35.6107,
            -30.5754],
           [-30.2230, -33.5731, -35.8306,  ..., -34.3179, -34.2813,
            -22.6298],
           [-15.9992, -27.0066, -34.1596,  ..., -34.1373, -32.0856,
            -19.3197]]]]], device='cuda:0'), None]

Is there any problem with this result?