Detectron2 taking too long on video processing time

[thiagoqd]

Hi,

Running "Run panoptic segmentation on a video" example on google colab with Tesla T4 GPU is taking too long.
A 5 minutes video with 30fps (~9k frames) has a estimated time of almost 6h processing:

Is that corret?
Running locally on my pc, with a GTX 1050ti (yeah, not that good), had 7h estimated time.

Video processing with Detectron2 take that long? Or may I'm doing something wrong?
Checking CUDA availability on Torch seems ok:

For a comercial prodution software whats the recommended GPU?

Thanks.