Object Detection Problem with Java API

Performance
,
1

Hello everyone,

I’m currently using JAVA API to implement the object detection using mxnet. The problem is the result of the object detection. If you guys see the java API example on the official tutorial, the machine detects all of the object inside the given picture (car, bicycle, and dog). After a couple of tries using different environment, mine could only detects the car (the result is: car,car,car). Anyone experiencing the issue ? What can I do to fix this behaviour. Fyi, I’m pretty sure that I’ve followed the tutorial correctly. Anyone with the solution other than repeating the tutorial ?

2

Hi,

do you mean that the bounding boxes around all the objects say car or that it is only able to detect (and draw a bounding box around the car). If it’s the former, then it’s like a problem with the data labels.

Could you post your code and a link to the example you’re following?

Are you following this example:

3

I’ve already following every steps carefully in this tutorial with no success :

Java Net inference tutorial with MXNET

And the exact problem is the bounding boxes is detecting the car repeatedly, so the other objects isn’t appeared because it only display 3 objects with the highest confidence level.

Any solution ?

4

Someone can give me a solution?

5

Hello, @kpratama24 and thanks for your interest in MXNet Java API!

I would like to see if I can be of help with your issue. I see that in your original question you mentioned:

After a couple of tries using different environment, mine could only detects the car (the result is: car,car,car).

In order to help you with your issue could you please provide a bit more details on what do you mean by “couple of tries” and specially “using different environment”. If you can outline exact steps you are following, even if they seem quite trivial to you, that would help to root cause your issue.

6

I mean using different operating system to see if it’s the OS issue itself.

7

Thanks for replying! It is still a bit unclear to me what steps lead you to detect a problem with classification. Would you be able to specify a list of actual steps one might need to follow to see the same result that you are seeing?

1 Like
8

Just follow every steps in the official mxnet java tutorial here:

mxnet Java Setup

After the setup is successful, just build and run the project and you will see that “car” is the only one that detected by the mxnet itself, even though the official documentation shows that the mxnet can detect 3 objects (car, dog, and bicycle). You can check all of them in the link above.

Looking forward into it.