lyh-18 / eqvi Goto Github PK

it seem that train_EQVI_lap_l1.py is used to train model with X4 interpolate, so what should I change if I want to train a model with X2 interpolate.

Hi: can you please provide a version of the code which works on windows using pycharm and cpu.

Hi. I am confused about your backwarp implementation. To my understanding, for a 2x2 image, the coordinates for the four pixels are (-0.5,-0.5), (-0.5, 0.5), (0.5, -0.5), (0.5, 0.5), respectively. But in your implementation, the coordinate of upper-left pixel is (0,0)? Did I misunderstand anything?

Hi,

Lots of interpolation method suffers from the subtitle issue, which in video like movie accompany with subtitle, while has great motion between two frame. It would cause subtitle area become blurry or complete damaged.

I wonder know whether your method could handle well in this case?

Thx,
Lei

Hello, your work is outstanding ！
But I didn’t understand the derivation of acceleration and velocity. You mentioned in the paper that if the directions of acceleration a1 and a2 are inconsistent, it will degenerate to the QVI model. If they are the same, then your model will be used.
My question is, why can’t I use the QVI model when the acceleration remains constant?

Hello: could you please provide the files to reproduce the training steps on gpu and cpu (for example for a small dataset).

could you add some script?

I failed to install correlation package

I see that your paper describes that the middle frame is inferred based on four frames, but there are currently only demos inferred from two frames on GitHub, how can I infer nonlinear motion based on the acceleration information of the object as mentioned in the paper?

Can you add sample interpolation results so that we can check the model is properly working on our system(by comparing PSNR/SSIM)? This should be helpful because different Pytorch and GCC versions might produce different results.

Hi,

I try to use the model in our repo. And I got some error as below:
return 20*output_dict_eval TypeError: unsupported operand type(s) for *: 'int' and 'dict'
I am curious about why 20 is multiply with output_dict_eval, which is not occurres in official repo of Scopflow .