I'm trying new 2019 version Openvino's Text Detection C++ demo. after some test, it works fine and outputs something like :
text detection model inference (ms) (fps): 655 1.52672
text detection postprocessing (ms) (fps): 141 7.0922
text recognition model inference (ms) (fps): 26.76 37.3692
text recognition postprocessing (ms) (fps): 0.0261467 38245.8
text crop (ms) (fps): 0.0584533 17107.7
However, it doesn't show any imshow window no matter how the -no_show tag is set or not set. Also, the -i tag cannot find a video file or camera.
what I have tried:
./text_detection_demo -i /dev/video0 -m_td ~/textmodel/text-detection-0002.xml -m_tr ~/textmodel/text-recognition-0012.xml -no_show 1
Could not read an image: /dev/video0
./text_detection_demo -i ~/a.png -m_td ~/textmodel/text-detection-0002.xml -m_tr ~/textmodel/text-recognition-0012.xml -no_show 0
works but no image show
./text_detection_demo -i ~/a.png -m_td ~/textmodel/text-detection-0002.xml -m_tr ~/textmodel/text-recognition-0012.xml -no_show 1
works but no image show
./text_detection_demo -i ~/a.png -m_td ~/textmodel/text-detection-0002.xml -m_tr ~/textmodel/text-recognition-0012.xml
works but no image shows ,neither
and if I want to read video file
./text_detection_demo -m_td ~/textmodel/text-detection-0002.xml -m_tr ~/textmodel/text-recognition-0012.xml -i ~/a.avi
Could not read an image: /home/gino/a.avi
what should I do?
Dear ininder, gg,
I'm assuming that you obtained your models from :
Correct ? I am looking at this issue for you. Thanks for your patience,
Hello ininder, gg,
How are you getting the text recognition results?
My understanding is you need a CTC decoder. I think there is an Intel OpenCL version of a greedy CTC decoder and possibly a CPU version too.
Have not checked the sample yet... maybe already using the CTC decoder?
What happens if you try
./text_detection_demo -i video.mp4 -m_td text-detection-0002.xml -m_tr text-recognition-0012.xml -dt video
Works fine here processing the input video and drawing text regions and OCR results
after taking a while look in text detection example,
I try CPU + MYRIAD ( I have only 1 NCS stick )
when using MYRIAD to detect, CPU to recognize they work fine as using CPU in both tasks, however
, if using CPU to detect, MYRIAD to recognize, it would find nothing and no text to be outputed.
I think it's the way the neural compute stick works.