Where are the codes for OmniScore computing and the overall training pipeline?

[QiushiYang]

It seems the repo only releases the codes for model inference and a part of model training. The core parts of overall training via DPO and the calculation process of OmniScore seems missing.

[TimZhang001]

It seems the repo only releases the codes for model inference and a part of model training. The core parts of overall training via DPO and the calculation process of OmniScore seems missing.

提供的数据集上已经有了paired的win和loss，根据win和loss应该可以进行训练。不过里面一些代码存在问题，需要进行修改

[CIntellifusion]

Thanks for your interest in our work. We are working heavily on clean the code and OmniScore Pipeline. It will be public as soon as possible. Sorry for causing trouble for your research progress.

[QiushiYang]

Thanks for your interest in our work. We are working heavily on clean the code and OmniScore Pipeline. It will be public as soon as possible. Sorry for causing trouble for your research progress.

Thank you for reply! Could you share any specific approaches (the used pre-trained models, usage manners, metrics) you used when calculating the OmniScores, for examle, the quality of temporal flickering, motion smoothness, and degree of motion dynamics towards inter-frame level?

[wsxwd]

+1

[STAR-CV]

I'd like to ask if the code for OmniScore-Based Data Re-Weighting method is open source ?