llmpass / rstt Goto Github PK

Hello,

MSU Graphics & Media Lab Video Group has recently launched two new Super-Resolution Benchmarks.

• Video Upscalers Benchmark: Quality Enhancement determines the best upscaling methods for increasing video resolution and improving visual quality.
• Super-Resolution for Video Compression benchmark aims to test Super-Resolution methods on compressed videos and select the best model for each video codec standard.

If you are interested in participating, you can add your algorithm following the submission steps:

• Submit for Video Upscalers Benchmark: Quality Enhancement
• Submit for Super-Resolution for Video Compression benchmark

We would be grateful for your feedback on our work!

Hello,Is the extra frame from the decoder just get the average of the adjacent features from the encoder?
if i % 2 == 0:
y[:, i, :, :, :] = x[:, i//2]
else:
y[:, i, :, :, :] = (x[:, i//2] + x[:, i//2 + 1]) / 2

Congratulations! very excellent work! But I have a small wonder.
I am a beginner in DL. This question may be stupid.
Shouldn't you call torch.cuda.synchronous() before start?
like this:

you may refer to this tarun005/FLAVR#14

In FLAVR, whether use torch.cuda.synchronous() leads to a significant difference in inference time.
their origin paper(before use this function):

and their current paper(after use this function):

Could you please report your comparison?

Looking forward to your reply. Thank you a lot!

If I change the number of input_nums, from input_frames=4 to input_frames=12, and cancel Interpolation ,will it affect the model results?
In addition, if it is feasible to convert to TensorRT?

Hi, came across your work by chance. Has this CVPR article been published yet?

how can someone test a video on these pre-trained models?

Hello,I am very interested in your work. I want to use this model to train my own dataset,is this possible?

Hello! Thank you for releasing the code of your amazing work!

I want to implement some experiments based on your pre-trained weight, but I think that we are not able to resume training from the released checkpoints without its corresponding ".state" file.

Could you please also release the .state files?

Thank you very much!

Hello, I trying to evaluate pre-trained models. But I can't find network configurations for RSTT-M and RSTT-L checkpoints. Can you provide it?

I trained my own dataset, but the effect of interpolation is not good. If I do not want to interpolation frames, that is, input_ frame=4,output_ frame=4, how should I modify it?
can i modify the output_frame in the weight of completed training?

Excuse me, I am training my own super resolution data. The HR_image size is 720 * 1080. What relevant parameter settings need to be changed?crop_size need to change the size to 720?

Excuse me, does frame insertion have a great impact on the output? Because my test results will have obvious flicker problems:For example, in the following output, the results of singular images are obviously better, and the results of even images are blurry