search

TRAILab / CaDDN

Categorical Depth Distribution Network for Monocular 3D Object Detection (CVPR 2021 Oral)
Apache License 2.0
359 stars 62 forks source link

Question: inference time #23

Closed frankfengdi closed 3 years ago

frankfengdi commented 3 years ago

Really nice work! Thank you for releasing the code. I wonder how large is the network in total, and how long does it take for inference?

codyreading commented 3 years ago

These numbers depend on the batch size and the hardware you are running inference with. I can give you these for my personal workstation for example:

GPU: NVIDIA Titan XP
Batch Size: 2
Max GPU Memory Consumption: 7712 MB
Avg. Iteration Speed: 1.82 it/s
Total Time for KITTI Val Inference: 17 min, 15 seconds

Feel free to run this yourself and see what kind of performance you get.

rockywind commented 3 years ago

It's very odd. My GPU is NVIDIA RTX3090, but it took 10s for every image. image

codyreading commented 3 years ago

Hi @rockywind,

Are you using torch==1.4.0? If you are using a newer torch version, the inference will take a very long time. If you want to keep your current environment with a newer torch version, you can change this line from:

            batch_dict[key] = kornia.image_to_tensor(val).float().cuda()

to:

            batch_dict[key] = kornia.image_to_tensor(val).float().cuda().contiguous()

I am working on changing this in the repo, but you can make this quick fix for now.

rockywind commented 3 years ago

Thanks a lot for your help. It's work.

codyreading commented 3 years ago

Great glad I can help