lowest performances with my own detect script for yolov5

[happylulux]

Hi,
I succeed to train a custom yolov5 model to detect tomato
I tried with the detect.py given by ultralytics and i'm happy with the result, but the code is so heavy and complicated to modify, i decided to test with the torch.hub.load to see the result.
My version of the detect script find very few tomato, but i use the same best_tomato.pt model...
Anyone have an idea of the problem?
here is the code:

# import the opencv library 
import cv2 
import numpy as np
import torch
  
# define a video capture object 
vid = cv2.VideoCapture(0) 
vid.set(cv2.CAP_PROP_FRAME_WIDTH, 640)
vid.set(cv2.CAP_PROP_FRAME_HEIGHT, 640)

model = torch.hub.load('ultralytics/yolov5', 'custom', path='best_tomato.pt', force_reload=False)


while(True): 
      
    # Capture the video frame 
    # by frame 
    ret, frame = vid.read() 
    #yolov5 prediction
    results = model(frame)  

    # Display the resulting frame 
    cv2.imshow('frame', frame) 

    print(results.pandas().xyxy[0])
    # the 'q' button is set as the 
    # quitting button you may use any 
    # desired button of your choice 
    if cv2.waitKey(1) & 0xFF == ord('q'): 
        break
  
# After the loop release the cap object 
vid.release() 
# Destroy all the windows 
cv2.destroyAllWindows()