I am trying to incorporate DART into my pipeline and am encountering a problematic performance fork:
- Distilled runs nicely but its results are not good enough.
- Pruned gives good results offline but OOMs the GPU.
I am wondering if there is a way to run Pruned in a more VRAM-economical way? Will reducing the --imgsz value help with this?
I am trying to incorporate DART into my pipeline and am encountering a problematic performance fork:
I am wondering if there is a way to run Pruned in a more VRAM-economical way? Will reducing the --imgsz value help with this?