inference example里面用的DiViDeAddEvaluator而不是TSAN，这是啥？

[hl0737]

如题，官方代码的inference.py里面用的是DiViDeAddEvaluator，这个在论文里面没提到

请问下TSAN的推理example有嘛？

为什么VTSS是论文里面的重大贡献，然后github里面放的example推理代码是DiViDeAddEvaluator呢？

[qiuqiuh]

Thank you for your question！

We trained several versions of the model for predicting VTSS. Since some scores are internal scores that are not readily available to the public, we released the model to predict scores simply by inputting the video.

We define TSAN to be a network capable of predicting VTSS and it is structure-independent. The appendix of the detailed results paper about the ablation of the network structure.

Thank you for your advice, we will modify the code to make it clearer for users.

[hl0737]

哥能说中文嘛😢我理解DiViDeAddEvaluator是个残血版的，满血版的TSAN考虑放出来嘛？感谢！

[hl0737]

噢，对了哥，想请教您一下，对于快慢放的视频如何筛选，您有什么经验么，尤其是那种极端快放的视频，光流不太行。。。

[HuaZheLei]

噢，对了哥，想请教您一下，对于快慢放的视频如何筛选，您有什么经验么，尤其是那种极端快放的视频，光流不太行。。。

Hi, as far as we know, there is currently no ready-made model for recognize fast or slow motion video. You could collect some video data, maybe about 1k, and train a VLM for recognition.

[hl0737]

噢，对了哥，想请教您一下，对于快慢放的视频如何筛选，您有什么经验么，尤其是那种极端快放的视频，光流不太行。。。

Hi, as far as we know, there is currently no ready-made model for recognize fast or slow motion video. You could collect some video data, maybe about 1k, and train a VLM for recognition.

Thanks for ur reply, 2025 fighting !