GPU memory of ZeroDepth

[baibizhe]

Hello. I am try to inference with
zerodepth_model = torch.hub.load("TRI-ML/vidar", "ZeroDepth", pretrained=True, trust_repo=True)
However , it is only possible if I resize input image to a extreme small size.For example, 144,256. If the image size is 640x360. There will be a OOM of GPU. I run all my experiments on A100 40G., Is this normal?

Best regards,
Bizhe

[mrussell9]

I'm seeing this same issue

Have you found a solution?

[baibizhe]

I'm seeing this same issue

Have you found a solution?

Not yet

[mrussell9]

I was able to run inference on a GPU with less VRAM than yours with

with torch.no_grad():
    depth_pred = zerodepth_model(rgb, intrinsics)