Bounding boxes are placed in the wrong coordinates

Community Help
1

I have some problem with the bounding boxes. The visualize model in Roboflow correctly predicts the objects from the image but when I try it with my own code, it doesn’t work with the models (27, 25) but it does work with models (11, 14).

The model in the browser works well. I tried running it with docker but it doesn’t work for the models (27, 25) but works with models (11, 14) as shown:

rf_1
rf_11920×1202 144 KB

I tried running it without docker and it works fine, same as the browser.

Here is my code:

import cv2
from roboflow import Roboflow
from inference_sdk import InferenceHTTPClient

client = InferenceHTTPClient(
    api_url="http://localhost:9001",
    api_key="################"
)

class ObjectDetection:
    def __init__(self):
        self.rf = Roboflow(api_key="################")
        #self.project = self.rf.workspace("stust-k-lab").project("kitchen-intelligent-assistant")
        #self.model = self.project.version(27).model
        self.main_dish_count = 0
        self.side_dish_count = 0
        self.width = 960
        self.height = 540

    def get_class_color(self, class_name):
        if "-m-" in class_name:
            self.main_dish_count += 1
            if self.main_dish_count == 1:
                return (0, 0, 255)  # Red
        elif "-s-" in class_name:
            self.side_dish_count += 1
            if self.side_dish_count == 1:
                return (0, 255, 0)  # Green
            elif self.side_dish_count == 2:
                return (255, 0, 0)  # Blue
            elif self.side_dish_count == 3:
                return (0, 255, 255)  # Yellow
        return (255, 255, 255)  # Default to white for unknown classes

    def process_yolov8(self, frame):
        frame = cv2.resize(frame, (self.width, self.height))
        #results = self.model.predict(frame, confidence=50, overlap=50)
        #predictions = results.json()
        with client.use_model(model_id="kitchen-intelligent-assistant/27"):
            predictions = client.infer(frame)
            
        print(predictions)
    
        for prediction in predictions['predictions']:
            x = prediction['x']
            y = prediction['y']
            width = prediction['width']
            height = prediction['height']
            class_name = prediction['class']
            confidence = prediction['confidence']
            x1 = int(x - (width / 2))
            y1 = int(y - (height / 2))
            x2 = int(x + (width / 2))
            y2 = int(y + (height / 2))
            color = self.get_class_color(class_name)
            cv2.rectangle(frame, (x1, y1), (x2, y2), color, 1)
            label = f'{class_name[:2].replace("-", "")} - {confidence:.2f}'
            text_size, _ = cv2.getTextSize(label, cv2.FONT_HERSHEY_SIMPLEX, 0.5, 1)
            cv2.rectangle(frame, (x1, y1 - text_size[1]), (x1 + text_size[0], y1), color, -1)
            cv2.putText(frame, label, (x1, y1), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 0, 0), 1)

        self.main_dish_count = 0
        self.side_dish_count = 0
        return frame

if __name__ == "__main__":
    yolo_processor = ObjectDetection()
    #video_file = "records\\record_20240312_135626.avi"
    #cap = cv2.VideoCapture(video_file)
    picture = cv2.imread('data\\0 degrees\day_1\lunch\day_1_lunch (1).png')
    while True:
        #ret, frame = cap.read()
        #if not ret:
        #    break
        
        predictions = yolo_processor.process_yolov8(picture)
        cv2.imshow("YOLOv8", predictions)

        if cv2.waitKey(int(1000/32)) & 0xFF == ord('q'):
                break
    
    cap.release()
    cv2.destroyAllWindows()
2

Nevermind, I figured out why. There was a problem with the preprocessing before training and re-trained the whole model again and it worked.

3

This topic was automatically closed 7 days after the last reply. New replies are no longer allowed.