There seems to be no effect in the road data

[LinkerRrrr]

@m1a0cX Hi,Thank you for your outstanding work.
I tried to use road data to remove vehicles on the road surface. I obtained vehicle masks through SAM3, but it seems that the vehicles cannot be removed. I want to know why this is the case?
This is the result
https://github.com/user-attachments/assets/4dfc604b-19bd-4399-bceb-6156fbb972d1

[m1a0cX]

I think there may be some failure cases. But I found the model totally failed to remove the car which is also out of my expectation. Could you provide the original video and the mask video so I can have a try?

[LinkerRrrr]

OK,here they are.Thanks for your reply!

https://github.com/user-attachments/assets/320e1e6a-b186-4af7-9c8b-dae8f9d0cf4e
https://github.com/user-attachments/assets/aa967de5-2812-4bbb-a342-d35dd98de82a

[m1a0cX]

OK,here they are.Thanks for your reply!

https://github.com/user-attachments/assets/320e1e6a-b186-4af7-9c8b-dae8f9d0cf4e https://github.com/user-attachments/assets/aa967de5-2812-4bbb-a342-d35dd98de82a

I see. I think our current model can't deal with too much masked objects in the video at the same time since our training process only contains single masked object. So when coming to your case the model fails to remove vehicles on the road. To improve the ability of removing multiple objects, I think it needs more training data which have this kind of training sample.

[LinkerRrrr]

Alright, then I also want to ask the same question: will the training code be made available?hahaha