Putting YoloV5 on Jetson Nano 2GB

[edwin.estep]

Hey all,

I’m trying to put yolov5 on the Jetson, but can’t get it to run. Specifically, I’m trying to use it with a CSI camera, which requires that the code be changed.

To install the yolov5 repo and dependencies, I entered commands from this guide:

git clone https://github.com/ultralytics/yolov5.git
cd yolov5
export OPENBLAS_CORETYPE=ARMV8
sudo apt install libfreetype6-dev python3-dev
sudo pip3 install numpy==1.19.4
sudo pip3 install --ignore-installed PyYAML>=5.3.1
sudo pip3 install -r requirements.txt

I updated the code according to these sources to try to get it to work:

NVIDIA-AI-IOT/jetcam/blob/master/jetcam/csi_camera.py

Yolov5 on jetson nano with csi-camera

Greetings! Has anyone run yolov5 on a jetson nano with a csi camera? Share your experience. How to run csi-camera in python on jetson nano?

Despite this, I’m getting an error. Here’s the stack trace:

MyProject@MyProject-desktop:~/Desktop/yolov5$ python3 detect.py --source=0
/usr/local/lib/python3.6/dist-packages/torchvision/io/image.py:11: UserWarning: Failed to load image Python extension: 
  warn(f"Failed to load image Python extension: {e}")
detect: weights=yolov5s.pt, source=0, data=data/coco128.yaml, imgsz=[640, 640], conf_thres=0.25, iou_thres=0.45, max_det=1000, device=, view_img=False, save_txt=False, save_conf=False, save_crop=False, nosave=False, classes=None, agnostic_nms=False, augment=False, visualize=False, update=False, project=runs/detect, name=exp, exist_ok=False, line_thickness=3, hide_labels=False, hide_conf=False, half=False, dnn=False
YOLOv5 🚀 v6.1-21-ge6e36aa torch 1.10.2 CPU

Fusing layers... 
Model Summary: 213 layers, 7225885 parameters, 0 gradients, 16.5 GFLOPs
[ WARN:0@27.567] global /io/opencv/modules/videoio/src/cap_v4l.cpp (1910) getProperty VIDEOIO(V4L2:/dev/video0): Unable to get camera FPS
[ WARN:0@37.616] global /io/opencv/modules/videoio/src/cap_v4l.cpp (1000) tryIoctl VIDEOIO(V4L2:/dev/video0): select() timeout.
1/1: 0...  Success (inf frames 3264x2464 at 99.00 FPS)

Traceback (most recent call last):
  File "detect.py", line 261, in <module>
    main(opt)
  File "detect.py", line 256, in main
    run(**vars(opt))
  File "/usr/local/lib/python3.6/dist-packages/torch/autograd/grad_mode.py", line 28, in decorate_context
    return func(*args, **kwargs)
  File "detect.py", line 109, in run
    dataset = LoadStreams(source, img_size=imgsz, stride=stride, auto=pt)
  File "/home/MyProject/Desktop/yolov5/utils/datasets.py", line 328, in __init__
    s = np.stack([letterbox(x, self.img_size, stride=self.stride, auto=self.auto)[0].shape for x in self.imgs])
  File "/home/MyProject/Desktop/yolov5/utils/datasets.py", line 328, in <listcomp>
    s = np.stack([letterbox(x, self.img_size, stride=self.stride, auto=self.auto)[0].shape for x in self.imgs])
  File "/home/MyProject/Desktop/yolov5/utils/augmentations.py", line 93, in letterbox
    shape = im.shape[:2]  # current shape [height, width]
AttributeError: 'NoneType' object has no attribute 'shape'

Does anyone know why this might be happening/have any experience running yolo on a Jetson (with a CSI Camera)?

[AastaLLL]

Hi,

First, since YOLOv5 is a relatively complicated model, Nano 2GiB may not have enough memory to deploy it.

The error is caused by the camera frame (im) being NULL.
So it seems some issue when reading the camera from OpenCV.

Would you mind checking if your camera is mounted at /dev/video0 first?
For example, below is the log from our device.
We have a CSI camera (ov5693) mounted at /dev/video0.

$ sudo apt-get install v4l-utils
$ v4l2-ctl --list-devices
vi-output, ov5693 2-0036 (platform:15c10000.vi:2):
        /dev/video0

Thanks.