Benchmark for inferencing time

[SamitM1]

Hey guys, great work here.
Wondering how long does it take to enhance a 720p video with heavy degradation to 2k resolution on an H100? Is it possible to use more gpus to make it faster? Anyway to make inferencing time under 90 or 120 secs?

Thanks in advance!

[CSRuiXie]

Hi, thanks for your interest! Regarding your questions:

1. We haven’t tested enhancing a 720p video with heavy degradation to 2k resolution on an H100 because we currently don’t have access to an H100 device. However, using our provided toy example (240x426), it takes about 20 minutes on an A100.
2. We plan to support multiple GPU inference. Once the code is ready, we will release it as soon as possible.
3. Currently, it’s not possible to accelerate the inference time to under 90 or 120 seconds. However, in the future, we may incorporate faster T2V models to achieve this.

[SamitM1]

@CSRuiXie
Thank you for your detailed response! My team is really impressed with your work, and we are highly interested in exploring potential partnership opportunities. We would also love to discuss how we can support your efforts, including providing access to GPUs and training data.

Could we possible connect via email?

[CSRuiXie]

Absolutely :D, here is my email: [email protected]

[SamitM1]

Awesome I sent an email.

[jiaxiangc]

@CSRuiXie
Hi大佬，这是一篇非常棒的工作。我有一些问题想请教：

1. 超分推理的时间是20分钟，这会大幅度削弱实用性。但是为什么推理时间在I2VGenXL这种结构的模型上还这么长，是因为分辨率的问题吗？
2. 为什么推理代码的视频分解要在bash进行，这是否会导致重复加载模型。