Running yolov5 on CPU

[riadsouissi]

Hi.
With such high fps with the small yolov5 model, would it run decently on a cpu configuration (xeon processor for instance)?

[github-actions]

Hello @riadsouissi, thank you for your interest in our work! Please visit our Custom Training Tutorial to get started, and see our Jupyter Notebook , Docker Image, and Google Cloud Quickstart Guide for example environments.

If this is a bug report, please provide screenshots and minimum viable code to reproduce your issue, otherwise we can not help you.

If this is a custom model or data training question, please note that Ultralytics does not provide free personal support. As a leader in vision ML and AI, we do offer professional consulting, from simple expert advice up to delivery of fully customized, end-to-end production solutions for our clients, such as:

• Cloud-based AI systems operating on hundreds of HD video streams in realtime.
• Edge AI integrated into custom iOS and Android apps for realtime 30 FPS video inference.
• Custom data training, hyperparameter evolution, and model exportation to any destination.

For more information please visit https://www.ultralytics.com.

[glenn-jocher]

@riadsouissi yes absolutely! I do all my development on a 2018 MacBook Pro (no CUDA devices), and if I run the webcam demo at the default --img-size 640 I get about 5 FPS on CPU:

(env1) glennjocher@Glenns-MBP yolov5 % python detect.py --source 0
Namespace(agnostic_nms=False, augment=False, classes=None, conf_thres=0.4, device='', fourcc='mp4v', half=False, img_size=640, iou_thres=0.5, output='inference/output', save_txt=False, source='0', view_img=False, weights='weights/yolov5s.pt')
Using CPU

1/1: 0...  success (1280x720 at 29.00 FPS).

0: 384x640 Done. (0.266s)
0: 384x640 1 persons, Done. (0.217s)
0: 384x640 1 persons, Done. (0.218s)
0: 384x640 1 persons, Done. (0.207s)
0: 384x640 1 persons, Done. (0.209s)
0: 384x640 1 persons, Done. (0.209s)
0: 384x640 1 persons, Done. (0.208s)
0: 384x640 1 persons, Done. (0.210s)
0: 384x640 1 persons, Done. (0.210s)
0: 384x640 1 persons, Done. (0.207s)

If I reduce to --img-size 320 I get about 15-20 FPS on CPU:

(env1) glennjocher@Glenns-MBP yolov5 % python detect.py --source 0 --img 320
Namespace(agnostic_nms=False, augment=False, classes=None, conf_thres=0.4, device='', fourcc='mp4v', half=False, img_size=320, iou_thres=0.5, output='inference/output', save_txt=False, source='0', view_img=False, weights='weights/yolov5s.pt')
Using CPU

1/1: 0...  success (1280x720 at 29.00 FPS).

0: 192x320 1 persons, Done. (0.096s)
0: 192x320 1 persons, Done. (0.064s)
0: 192x320 1 persons, Done. (0.064s)
0: 192x320 1 persons, Done. (0.058s)
0: 192x320 1 persons, Done. (0.062s)
0: 192x320 1 persons, Done. (0.063s)
0: 192x320 1 persons, Done. (0.062s)
0: 192x320 1 persons, Done. (0.061s)

If you download our iDetection app, you will get 30+ FPS also on the latest iPhones. All of Apple's recent mobile devices (iPads, iPhones, etc) actually run YOLOv5 much faster than Intel CPU devices like MacBooks. This is due to the ANE, which for the last two years has had about 5 TOPS of power, and is likely due for a significant boost with the fab process reduction this year to 5nm.

[riadsouissi]

Thanks for the prompt feedback!

[rkinas]

Hi,
BTW does iDetection app support y5? I just downloaded it on iPad Pro and can see only y4.

[glenn-jocher]

@rkinas ah, yes. This is on our TODO list. Unfortunately we are stretched pretty thin, but probably by next week we will have new y5 models exported to iDetection.

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[themrmax]

If anyone ends up here because train.py is hanging, I had to use --workers 0 to get it to run on my 2017 Macbook Pro, I thinks it's related to this issue pytorch/pytorch#1355