How to run inference.py: missing jin_et_al files and pretrained_Blur2Vid.pth

[alhahope]

Hi, thanks for releasing the code.

We are trying to use HyperCUT as a baseline for recovering latent sharp subframes from a single motion-blurred image, but we ran into the same issue discussed here.

Specifically, inference.py with --backbone Jin seems to require:

1. models/backbones/jin_et_al/*
2. pretrained_Blur2Vid.pth

However, the current repository does not include models/backbones/jin_et_al, and the README appears to provide the HyperCUT ordering checkpoint rather than the full Blur2Vid generation checkpoint expected by inference.py.

We noticed that downloading the original Jin et al. repository can bypass the missing import, but it is still unclear which checkpoint is actually compatible with HyperCUT's inference.py. Using Jin checkpoints such as center_v3.pth does not seem equivalent to the expected pretrained_Blur2Vid.pth.

Could you clarify:

• Is pretrained_Blur2Vid.pth publicly available?
• What exact directory structure is expected for models/backbones/jin_et_al?
• Which checkpoint should be used for single-image inference to generate the 7 sharp frames?
• Is inference.py intended to run the full blur-to-video generation pipeline, or only the Blur2Vid backbone before HyperCUT ordering?

A minimal working inference command and checkpoint link would be very helpful.

Thank you!