Ask Your Question
0

SSD_mobileNet displaying N/A 99%

asked 2017-11-07 07:56:16 -0600

Adez gravatar image

Good day guys. I just trained SSD_mobileNet on two classes. I used LabelImg to draw bounding box and label my dataset. After training I froze my .pbtxt file to get .pb file. I used tensorflow object_detection.py code with openCV. During my testing on my laptop, I got the a bounding box with N/A 99% on it and it did not recognise any on the two classes I trained with. I dont know what to do. Please help me guys.

edit retag flag offensive close merge delete

Comments

Hey, assuming you used the following code to load your network:

net = cv.dnn.readNetFromTensorflow(prototxt, weights)

what did you use for the 'prototxt' file?

I also retrained MobilenetSSD, using the Tensorflow Object Detection API, and then used the prototxt file from here https://github.com/opencv/opencv_extr... modified to the number of classes of my retrained network. However, my output is very bad with boxes all over the place. My theory Is that there is a mismatch between the structure of config file used in the Object Detection API and the structure used by OpenCV dnn.

Any suggestions?

XenonHawk gravatar imageXenonHawk ( 2018-01-12 09:33:07 -0600 )edit

1 answer

Sort by ยป oldest newest most voted
2

answered 2017-11-10 10:41:03 -0600

Adez gravatar image

I have solved the question. My Label_map format was wrong, I omitted display_name in my Label_map. Thank you guys.

edit flag offensive delete link more

Comments

Opencv 3.3.0 or 3.3.1?

Robb gravatar imageRobb ( 2017-11-10 11:19:52 -0600 )edit

version 3.10

Adez gravatar imageAdez ( 2017-11-17 01:11:08 -0600 )edit

Question Tools

1 follower

Stats

Asked: 2017-11-07 07:56:16 -0600

Seen: 845 times

Last updated: Nov 10 '17