Couldn't display YOLO result via OpenCV dnn module

[barzan-hayati]

@AlexeyAB
I'm trying to train YOLO object detection via darknet and it was great for showing object via ./darknet detector test cfg/obj.data cfg/yolo-obj.cfg yolo.weights data/test.jpg
I used 160*30 images for train and test. Also I did it via Tiny-YOLOv2 and another model and it could detect all numbers in the images.

But when I want to showing result for trained weight and a specific image like bellow, and by using Opencv dnn modules, I could not display all objects correctly.

YOLO Object Detection with OpenCV and Python
Deep Learning based Object Detection using YOLOv3 with OpenCV ( Python / C++)
YOLO object detection with OpenCV
YOLO object detection using OpenCV and Python
Exploring OpenCV’s Deep Learning Object Detection Library

But when I check it for a bigger image(1280*760), It could show results.

[barzan-hayati]

Correct output:

[barzan-hayati]

correct output for plate detection:

[AlexeyAB]

There are different approaches of resizing in Darknet and OpenCV-dnn: #232 (comment)

Try to Request feature, to use simple Resizing or Letter_box in the OpenCV-dnn module, there: https://github.com/opencv/opencv/issues

[barzan-hayati]

@AlexeyAB Thanks for your respond.
I solved it. When we want to showing result using opencv and dnn module, we must consider ratio of resizing of original image. For example when I have a 160*30 input image, at first I convert it to 416*78(resizing one of the width or height to 416). Now for showing results, I resize height by multiplying it to 416/78