tenzin

但是那些好像都是在coco数据集上训练的，不能跑voc的数据集

Thanks for your response. So if i want to reproduce the result on Distinctions-646 reported in paper, i need to replace the [vit.py](https://github.com/hustvl/ViTMatte/blob/main/modeling/backbone/vit.py) in ViTMatte with the [vit.py](https://github.com/hustvl/Matte-Anything/blob/main/modeling/backbone/vit.py) in MatteAnything?

Thanks for your reminding. I was curious whether the pseudo trimap in MatteAnything(Table 1 2 3 4) was obtained through a real user study or a simulation of user interactions...