I’m trying to do a system where I have:

• a detector based on YOLOv8 and trained on my custom dataset
• a classifier that can be for example EfficentNet
  I have already trained these two and am now testing the system on a video to see the results.
  The detector is been trained with:

model = YOLO(model_path)

results_m = model.train(data= data, single_cls=False, imgsz=640, epochs = 25) 

The problem I am encountering is that the detection phase seems to fail on static objects. Specifically, my dataset has 3 classes: plastic bags, cardboard boxes, and an “other” class. If a plastic bag is carried by a person, the system detects it, but if the plastic bag is dumped and then stops on the ground, the system no longer detects it. Why does this happen?

My code is the following:

from tensorflow.keras.models import load_model
import cv2
from ultralytics import YOLO

classifier_class_names = {
    0: 'Cardboard_box',
    1: 'Other',
    2: 'Plastic_bag',
}

detector = YOLO("/myPath/best.pt")

classifier = load_model("/myPath/efficientnet_model_unfreeze_128.h5")

video_path = "/myPath/video_test.mp4"
cap = cv2.VideoCapture(video_path)

output_path = "/myPath/video_out.mp4"
fourcc = cv2.VideoWriter_fourcc(*'mp4v')
fps = cap.get(cv2.CAP_PROP_FPS)
width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
out = cv2.VideoWriter(output_path, fourcc, fps, (width, height))

while cap.isOpened():
    ret, frame = cap.read()
    if not ret:
        break

    results = detector(frame)         
    detections = results[0].boxes   #list of boxes     

    for box in detections:
        x1, y1, x2, y2 = box.xyxy[0].tolist()
        conf = box.conf[0].item()
        cls = results[0].names[box.cls[0].item()]   

        # Extract ROI
        roi = frame[int(y1):int(y2), int(x1):int(x2)]

        # Preprocess ROI
        roi_resized = cv2.resize(roi, (300, 300)) 
        roi_resized = roi_resized / 255.0  
        roi_resized = roi_resized.reshape(1, 300, 300, 3)

        # Classify ROI
        pred = classifier.predict(roi_resized)
        class_id = pred.argmax(axis=1)[0]
        class_name = classifier_class_names.get(class_id, 'Other')

        label = f'Class: {class_name}, Conf: {conf:.2f}'
        cv2.rectangle(frame, (int(x1), int(y1)), (int(x2), int(y2)), (0, 255, 0), 2)
        cv2.putText(frame, label, (int(x1), int(y1) - 10), cv2.FONT_HERSHEY_SIMPLEX, 0.9, (0, 255, 0), 2)
    
    out.write(frame)

cap.release()
out.release()
cv2.destroyAllWindows()