I am using a custom model based on tiny-yolov3 on the openvino framework. while performing inference on the openvino platform, i'm getting the size of detected objects i.e (the bounding boxes) are somewhat bigger in size. I tested the same model on opencv dnn library and found that opencv is giving the exact bounding boxes. i am pretty much confused over this issue and cannot figure out the exact issue behind this problem. For your information i am using using openvino 2019 R1.
Dear Chakraborty, Subhasis,
Have you root caused this issue yet ? It is indeed strange that openvino is giving you slightly too large bounding boxes while opencv is giving you perfectly sized ones.
Which code in openvino are you using to perform inference - the object_detection_demo_yolov3_async sample (either Python or C++) ?