howiema / deepsort_yolov5_pytorch Goto Github PK

Hi there. I have tried running the programme on 2 seperate instances and have encountered these errors. I have made sure to have connected the webcam (the only webcam too), so that it takes --cam 0 correctly.

Here are the terminal outputs

python3 main.py --cam 0 --display
Namespace(agnostic_nms=False, augment=False, cam=0, classes=[0], conf_thres=0.5, config_deepsort='./configs/deep_sort.yaml', device='', display=True, display_height=600, display_width=800, fourcc='mp4v', frame_interval=2, img_size=640, input_path='input_480.mp4', iou_thres=0.5, save_path='output/', save_txt='output/predict/', weights='yolov5/weights/yolov5s.pt')
Initialize DeepSORT & YOLO-V5
Using CPU

Using webcam 0
Done..
Camera ...
Done. Create output file  output/results.mp4
Illegal instruction (core dumped)

And in another instance, I have a different error citing camera error

python3 main.py --cam 0 --display
Namespace(agnostic_nms=False, augment=False, cam=0, classes=[0], conf_thres=0.5, config_deepsort='./configs/deep_sort.yaml', device='', display=True, display_height=600, display_width=800, fourcc='mp4v', frame_interval=2, img_size=640, input_path='input_480.mp4', iou_thres=0.5, save_path='output/', save_txt='output/predict/', weights='yolov5/weights/yolov5s.pt')
Initialize DeepSORT & YOLO-V5
Using CPU

Using webcam 0
Done..
Camera ...
[ WARN:0] global /tmp/pip-req-build-tpkxoqhj/opencv/modules/videoio/src/cap_v4l.cpp (1001) tryIoctl VIDEOIO(V4L2:/dev/video0): select() timeout.
Traceback (most recent call last):
  File "main.py", line 259, in <module>
    with VideoTracker(args) as vdo_trk:
  File "main.py", line 75, in __enter__
    assert ret, "Error: Camera error"
AssertionError: Error: Camera error

The camera error occured right after the illegal instruction (core dumped) episode. When the programme was first run, the windows for the display could be open, but there is no video feed, and it just stops responding in Ubuntu (window goes greyscale)

Error title is :

python3.8 crashed with SIGILL in exec_blas()

If required, my device's techncial specs
DJI Manifold 2G (Nvidia Jetson TX2)
ARMv8 Processor rev 3 (v8l) × 4 ARMv8 Processor rev 0 (v8l) × 2
NVIDIA Tegra X2 (nvgpu)/integrated
8GB ram, Ubuntu 18.04 LTS

As for the new model of yolov5, I have checked that the version of yolov5 you used is 2.0, so I trained a helmet detection model with the corresponding version, and it has been tested by yolov5 alone and is effective. However, after I imported this model into your project, nothing can be detected. Is there anything that needs to be modified.

Looking forward to your reply very much!

1.找到选择device的区域

发现没有进入if语句，查看if语句之前的device.type确定是cpu，但是if判断语句中’cpu’是字符串，而device.type是cpu <class 'torch.device'>所以没有进入if语句。

2.device.type的类别是字符串，修改为device.type。还是报错

分析：追溯是执行vdo_trk.run()是出现问题。查看self.image_track(img0)使用的设备，发现是cpu，没问题。接着查看time_synchronized,发现错误

/home/ahei/DeepSORT_YOLOv5_Pytorch-master/yolov5/utils/torch_utils.py line 53 time_synchronized

原来，首先检查了torch.cuda.is_available()有的话调用了torch.cuda.synchronize()，然而我们的gpu在跑代码显存满了，所以报错。修改如下，问题解决。

Hello, we run the code successfully, but we find a question of the time costing.
In the sort part, the average time is about 40ms, but there are few frames taking more than 500ms. After investigation, it is determined that the time-consuming part of network (feature extraction part) reasoning has increased sharply, and the Layer4 takes the most time.
Have you encountered any similar problems? How to solve them, thanks!

Environment:
Xavier AGX 16GB - Jetpack 4.4
python 3.6.9
pytorch 1.6.0
opencv 4.1.1

Hi,I successfully run this project,but I have tried a lot of weights(yolov5s.pt) of different version from yolov5 official website,It kept telling me with"Can't get attribute 'C3' on <module 'models.common' >"So which version of yolov5 should I choose?Thank you so much.

Hello. I am trying to run your programme after installing pip install -r requirements.txt
However, i have ran into an error "TypeError: load() missing 1 required positional argument: 'Loader'"

This is the full output:

$ python main.py --cam 0 --display
Namespace(agnostic_nms=False, augment=False, cam=0, classes=[0], conf_thres=0.5, config_deepsort='./configs/deep_sort.yaml', device='', display=True, display_height=600, display_width=800, fourcc='mp4v', frame_interval=2, img_size=640, input_path='input_480.mp4', iou_thres=0.5, save_path='output/', save_txt='output/predict/', weights='yolov5/weights/yolov5s.pt')
Initialize DeepSORT & YOLO-V5
Using CPU

Using webcam 0
Traceback (most recent call last):
  File "main.py", line 259, in <module>
    with VideoTracker(args) as vdo_trk:
  File "main.py", line 53, in __init__
    cfg.merge_from_file(args.config_deepsort)
  File "/home/dji/Desktop/targetTrackers/howieMa/DeepSORT_YOLOv5_Pytorch/utils_ds/parser.py", line 23, in merge_from_file
    self.update(yaml.load(fo.read()))
TypeError: load() missing 1 required positional argument: 'Loader'

I have found some suggestions on github, such as in here https://stackoverflow.com/questions/69564817/typeerror-load-missing-1-required-positional-argument-loader-in-google-col,
which suggests to change yaml.load to yaml.safe_load

However, doing so leads me to this error instead

$ python main.py --cam 0 --display
Namespace(agnostic_nms=False, augment=False, cam=0, classes=[0], conf_thres=0.5, config_deepsort='./configs/deep_sort.yaml', device='', display=True, display_height=600, display_width=800, fourcc='mp4v', frame_interval=2, img_size=640, input_path='input_480.mp4', iou_thres=0.5, save_path='output/', save_txt='output/predict/', weights='yolov5/weights/yolov5s.pt')
Initialize DeepSORT & YOLO-V5
Using CPU

Using webcam 0
Done..
Camera ...
Done. Create output file  output/results.mp4
Illegal instruction (core dumped)

Has anyone encountered anything similar ? Thank you !

I would like to connect with you ...

Also, have a look at
https://github.com/mahimairaja/vehicle-counting-yolov5

Hi, I click the link of yolov5 weights, it seems in a google drive trash.
But then I realized it must be an official weight offered by glenn-jocher.
So, maybe you can change the link to a workable one or link to the yolov5?

Hello, I want to detect only some specific label in video. Where is the label for tracking?

Hey,wonderful job! But when I used the latest weights , error shows like "AttributeError: Can't get attribute 'C3' on <module 'models.common' ".And I found out that latest yolov5 project is necessary. Do you have any plan to update this project?

Hello, I am unable to run this implementation

I want to draw trajectory line for moving object through the video
how to do that?