Replies: 5 comments
-
Have you seen the same problem with Predictor API? The use case may not fit the ObjectDetector for you. You may need to write your personal post processing toolkit. |
Beta Was this translation helpful? Give feedback.
-
Thanks for the response. It's working fine with Predictor API. I have an arrays of output 1.0's with length 512. My question is, how would I apply the fix to Object Detection? What you mean by post processing toolkit is that converting those 1.0's to an image with prediction and everything related to the detection(probability, class)? |
Beta Was this translation helpful? Give feedback.
-
@lanking520 could you please give me an example of post processing tookit? Thanks. |
Beta Was this translation helpful? Give feedback.
-
I mean you may need to resolve your personal use cases. The predictor's result should bring you some bounding box probabilities and coordinates of the box in the image. To assist you better, I would like to know what does it mean to the output. If it is the same model, I remember there are three NDArrays output. What are they representing? |
Beta Was this translation helpful? Give feedback.
-
Thanks for the reply @lanking520, actually yesterday, I've tried this in my code: and it works fine with an output of:
Am thinking that maybe part of the code I used, I could pull the coordinates of the box in the image. Yes there are 3 NDArrays output: Thanks. |
Beta Was this translation helpful? Give feedback.
-
Hi @lanking520. After pulling your changes for the fix (#14804) of my issue, I tried it right away but am having now a strange error using the example of ObjectDetection.java class but I made some alteration to it to meet my requirements. Now it looks like this:
https://gist.github.com/androuino/7808b6fdf05e3122a03f35c63d3a5f89
I followed the step by step tutorial here https://github.com/apache/incubator-mxnet/tree/master/scala-package/mxnet-demo/java-demo for running the demo.
and this is the error am having:
The image I was used to test is 512x512 in size and the model that I've trained is also set to 512.
This is my train_yolo3.py script: https://gist.github.com/androuino/af5212923534a204b155c01b3bacb7f1
Then run the script with this command python train.py --gpus 0 --batch-size 2 --data-shape 512
"If you want the files that am using including the test image, I could email it you. I couldn't upload it publicly so I preferred to email it directly to you".
Thanks for any enlightenment that you could give me with this error or at least tell me if I did something or missed something with the code: https://gist.github.com/androuino/7808b6fdf05e3122a03f35c63d3a5f89 .
Beta Was this translation helpful? Give feedback.
All reactions