MGIF dataset reconstruction result not good

[JialeTao]

Hi, I use the provided checkpoint of MGIF dataset to run evaluation code for the "mode" reconstruction, and get the L1 metric of 0.027 while the paper report the 0.0206. There is a big difference between them. So I doubt that if I make sth. wrong. I use the pytorch version 1.4.0.

[AliaksandrSiarohin]

Can you post some of your visual results here?

[JialeTao]

Thanks for quick reply. I make a screen shot of the reconstructed gif as follow:

[AliaksandrSiarohin]

Can you send an entire gif?

[JialeTao]

Hi, I uploaded a few results.

https://user-images.githubusercontent.com/38600167/130410442-89c6ba07-2b14-499e-9007-bd37f9b25a1b.mp4

https://user-images.githubusercontent.com/38600167/130410472-b06bf071-48f3-4530-9337-12d2600fb97e.mp4

https://user-images.githubusercontent.com/38600167/130410490-1ef742d7-6acf-4601-9dd6-d7a00edab835.mp4

https://user-images.githubusercontent.com/38600167/130410525-54eed57f-1d74-48c6-905f-b2ed927a5513.mp4

https://user-images.githubusercontent.com/38600167/130410666-b673c401-6aac-4d23-87e5-005ce7869dc9.mp4

[AliaksandrSiarohin]

I see, so there are some problematic gray artifacts on the border. I need to check it out.

[JialeTao]

Yeah, I just noticed that. Thanks for checking.

[JialeTao]

I see, so there are some problematic gray artifacts on the border. I need to check it out.

Hi, is the cause of this phenomenon checked out?