For car plate detection repo, is that possible input only 1 image instead of video stream, and see the pop up detection results?


TensorRT Version :
CUDA Version :
Operating System + Version :
Jetson NX (jetpack 4.6)

I just followed the steps from the repo:

How should I do to input only 1 image instead of video stream, and see the pop up detection results?
I already tried

python3 --model=./networks/az_plate/az_plate_ssdmobilenetv1.onnx --class_labels=./networks/az_plate/labels.txt --input_blob=input_0 --output_cvg=scores --output_bbox=boxes /data/images out.jpg

got the message

[image]  loaded '/home/edward/Desktop/yolov4-custom-functions-master/data/images/car2.jpg'  (1400x950, 3 channels)
detected 0 objects in image
[OpenGL] glDisplay -- set the window size to 1400x950
[OpenGL] creating 1400x950 texture (GL_RGB8 format, 3990000 bytes)
[cuda]   registered openGL texture for interop access (1400x950, GL_RGB8, 3990000 bytes)
[image]  saved 'out.jpg'  (1400x950, 3 channels)

[TRT]    ------------------------------------------------
[TRT]    Timing Report ./networks/az_plate/az_plate_ssdmobilenetv1.onnx
[TRT]    ------------------------------------------------
[TRT]    Pre-Process   CPU   0.07171ms  CUDA   0.22339ms
[TRT]    Network       CPU  25.94165ms  CUDA  25.20064ms
[TRT]    Post-Process  CPU   0.14458ms  CUDA   0.14438ms
[TRT]    Total         CPU  26.15794ms  CUDA  25.56842ms
[TRT]    ------------------------------------------------

[TRT]    note -- when processing a single image, run 'sudo jetson_clocks' before
                to disable DVFS for more accurate profiling/timing measurements

and there is a result image out.jpg without any bounding boxes on it.
any idea on this?


You can update the into image input and output.
Below is an example for your reference:


Hi @AastaLLL,

Thanks for the response. I am wondering why It can detect the car plate correctly when I use webcam to point to my screen, which only shows the car image. However it cannot detect anything(the message shows: detected 0 objects in image) when I change the input from video to the car image.


Could you check if this happens to a specific image or every car image?

If only for a certain image, it should be the limitation of the model.
If it doesn’t work on each image input, could you share the detailed steps/source to reproduce this?