Object detection API return incorrect bounding box data?

Describe your question/issue here! (delete this when you post)

Here is how you can reproduce it:

  1. go to the timber log project: victor-manuel-opazo-troncoso-cyl8e/timber-log-v1
  2. open the network inspect from browser, and filter the calls with keyword “timber-log-v1”.
  3. drag and drop this image for OD prediction (https://static.vecteezy.com/system/resources/previews/031/318/703/non_2x/a-large-circular-piece-of-wood-wooden-trunk-or-stacked-tree-lumber-for-furniture-industry-wooden-log-concept-by-ai-generated-free-photo.jpg)


You can see that from my screenshot, the coordinate of one of box is one of the range, 475.5(x) + 67.0(width) is already out of width of the image which is 512. I cannot use it in my app to draw bounding boxes.

However, the roboflow UI shows correctly!!! Is there additional transformation needed?

I already post everything needed, including the project, the error and the OS/Browser :slight_smile:

Hi @Jack

The response format for a object detection model on our Hosted API, as it’s published on the docs, has the properties x and y which represent the center point of the detection, as well as the width and height. So, the right side of your detected bounding box should be 475.5 + 33.5 = 509, which is within you image size.

Hope this helps

ohhh, my apology!! Thank you Stella!

1 Like

This topic was automatically closed 7 days after the last reply. New replies are no longer allowed.