cv2.error: OpenCV(4.9.0) 👎 error: (-5:Bad argument) in function 'rectangle'

[topherbuckley]

Getting the following error on calls to results.save() where results is an instance of <class 'yolov5.models.common.Detections'>

yolov5-pip:

-> results.save()
(Pdb) n
cv2.error: OpenCV(4.9.0) 👎 error: (-5:Bad argument) in function 'rectangle'

Overload resolution failed:

• img marked as output argument, but provided NumPy array marked as readonly
• Expected Ptrcv::UMat for argument 'img'
• argument for rectangle() given by name ('thickness') and position (4)
• argument for rectangle() given by name ('thickness') and position (4)
  /media/HDD/git/beach-cleaning-robots/object-detection/scripts/model_evaluate.py(29)model_evaluate()

To reproduce:

1. pip install yolov5
2. run python model_evaluate.py best.pt ./dataset/test/_annotations.coco.json where model_evaluate.py is given below and best.pt is a local copy of a finetuned model:

from globox import AnnotationSet
import sys
import yolov5
# import torch


def model_evaluate(model_path, gt_path):
    # load ground truth dataset
    gt = AnnotationSet.from_coco(gt_path)
    images = ["dataset/test/" + item for item in list(gt.image_ids)]

    # Load model
    model = yolov5.load(model_path)
    # model = torch.hub.load("ultralytics/yolov5", "custom", path=model_path)

    # set model parameters
    model.conf = 0.25  # NMS confidence threshold
    model.iou = 0.45  # NMS IoU threshold
    model.agnostic = False  # NMS class-agnostic
    model.multi_label = False  # NMS multiple labels per box
    model.max_det = 1000  # maximum number of detections per image

    ## Run inference on all images within dataset
    results = model(images)
    # Single image
    breakpoint()
    results.save()
    results.pandas().xyxy[0].to_json(path_or_buf="output", orient="records")

    ## Save predictions to file

    ## Initialize AnnotationSet for detections


if __name__ == "__main__":
    model_path = sys.argv[1]
    gt_path = sys.argv[2]
    model_evaluate(model_path, gt_path)

I note I don't see this when using the code directly from the yolov5 git repo (as opposed to this pip package).

Original yolov5 repo:

-> results.save()
(Pdb) n
Saved 31 images to runs/detect/exp8
> /media/HDD/git/beach-cleaning-robots/object-detection/scripts/model_evaluate.py(29)model_evaluate()

Steps to reproduce:
3. pip uninstall yolov5
4. Swap the following code below and run PYTHONPATH=/home/bo/git/yolov5 python model_evaluate.py best.pt ./dataset/test/_annotations.coco.json:

#import yolov5
import torch
#model = yolov5.load(model_path)
model = torch.hub.load("ultralytics/yolov5", "custom", path=model_path)