DetecNet live inference without confidence in bounding box

[wannn-one]

Hi everyone,

I’m working on a live inference project using DetectNet with a custom SSD-MobileNet model on a Jetson device. My goal is to perform object detection and display bounding boxes without showing the confidence score within the bounding boxes.

Here’s what I’ve done so far:

• I’m using jetson-inference to load the model and detect objects.
• I tried using OverlayText to label the detected objects, but it always shows the confidence score, which I don't need.
• I also tried directly drawing bounding boxes without any text, but I still want to label the detected classes (like "helmet", "no_helmet", etc.) without the extra confidence information.

Here’s a simplified version of my code:

from jetson_inference import detectNet
from jetson_utils import videoSource, videoOutput

# Load the model
net = detectNet(model="model/ssd-mobilenet-300e.onnx", labels="model/labels.txt", 
                 input_blob="input_0", output_cvg="scores", output_bbox="boxes", 
                 threshold=0.5)

# Initialize video source and display
camera = videoSource("/dev/video0")  # For real-time camera input
display = videoOutput("display://0") # For display output

while display.IsStreaming():
    img = camera.Capture()

    if img is None:  # Capture timeout
        continue

    # Detect objects
    detections = net.Detect(img)
    
    # Reset counts for this frame
    frame_counts = {
        "helmet": 0,
        "no_helmet": 0,
        "vest": 0,
        "no_vest": 0
    }
    
    # Count objects in the current frame
    for detection in detections:
        class_id = detection.ClassID
        if class_id in class_labels:
            label = class_labels[class_id]
            frame_counts[label] += 1
    
    display.Render(img)
    display.SetStatus(f"Detection | FPS: {net.GetNetworkFPS()}")

I also tried to change output_cvg to "coverage" and an error just occured

[TRT]    3: Cannot find binding of given name: coverage
[TRT]    failed to find requested output layer coverage in network
[TRT]    device GPU, failed to create resources for CUDA engine
[TRT]    failed to create TensorRT engine for model/ssd-mobilenet-300e.onnx, device GPU
[TRT]    detectNet -- failed to initialize.

Could anyone guide me on how to:

• Draw bounding boxes with just class labels (e.g., "helmet") without the confidence score?
• Avoid using OverlayText if it forces confidence display or any alternative solution to this?
  Any help or suggestions would be greatly appreciated!

Thanks in advance!