Forums

NTIRE 2021 Video Super-Resolution Challenge - Track2. Spatio-Temporal Forum

Go back to competition Back to thread list Post in this thread

> why need the odd frame for the test dataset?

Hi everyone,
Could someone tell me why the odd frames of test dataset are needed? As I understand, we only need the LR even frames to predict HR whole frames, The inference procedure is: we pick up the LR even frames 00000000.png, 00000002.png, 00000004.png,.....00000098.png as input, and get output 00000000.png, 00000001.png, 00000002.png,.....,00000099.png, Do I understand correctly?
I am so confused about this, Could someone give me some advices?
Thanks soooo much.

Posted by: zhangqi @ Jan. 11, 2021, 4:05 a.m.

Hi zhangqi,

As you said, only the even-numbered frames are the inputs at inference(testing) time.
I provided both the even and the odd-numbered frames as they have already been public since NTIRE 2019 challenges.
I will update the data section regarding the inputs to clarify the policy.

Thanks for your feedback.

Best,
Seungjun

Posted by: SeungjunNah @ Jan. 11, 2021, 6:01 a.m.
Post in this thread