detection.py shows video but NO detections on file input (works fine on /dev/video0)

[jannskiee]

I am encountering an issue running the standard detection.py example on a Raspberry Pi 5 with the Hailo AI Kit.

• Webcam Source (/dev/video0): Detection works perfectly. Bounding boxes and labels are drawn correctly.
• Video File Source (.mp4): The video plays smoothly in the window, but zero detections are shown (no bounding boxes, no labels). The FPS counter updates, but the overlay remains empty.

To Reproduce

1. Setup Raspberry Pi 5 (8GB) with Hailo-8 AI Kit.
2. Use a custom YOLOv11s HEF model (verified working on webcam) and corresponding JSON labels.
3. Run the following command pointing to a local MP4 file:

python hailo_apps/python/pipeline_apps/detection/detection.py \
    --hef-path ~/Desktop/hef/yolov11s.hef \
    --input /home/admin/Desktop/hef/vid-3.mp4 \
    --labels-json ~/Desktop/hef/vehicles.json \
    --show-fps

Expected Behavior
The video file should play with bounding boxes drawn around detected vehicles, matching the behavior seen when using the webcam input.

Actual Behavior
The video playback works, but no inference results (bounding boxes) are displayed.

System Configuration

• Hardware: Raspberry Pi 5 + Hailo AI Kit (Hailo-8L)
• OS: Raspberry Pi OS 64-bit
• Hailo Software: hailo-all (installed via apt)
• Repository: hailo-apps (latest pull)
• Model: Custom YOLOv11s (compiled to HEF)

[hailocs]

@jannskiee
Does simply hailo-detect work? By default it takes an .mp4 sample from hailo-apps resources.

[jannskiee]

yes, the hailo-detect works it display the video and the detection however, using my own mp4 with my own hef file only shows the video with no detection even though when using my webcam as an input with the same hef file shows the video and detection.

note: it is a vehicle type detection hef file, the video includes vehicles and should show detections more accurately than using my webcam playing the video in my screen and letting the webcam capture it