The log images seems to be cropped or smaller in size although bounding box is covering whole object

I have annotated the data taking care of the bounding box ie the bounding box must cover the whole object. After implementing the model, although the bounding box is covering the FOV of the object the log’s image I am getting is smaller in size. Sometimes it is half of the particular object, sometimes is 75% of the object, and sometimes even it is the desired logs that I am expecting. No such similar pattern is found.
I am expecting the image of the logs should be the same as in the bounding box.
I am using deep stream 5.1.

I am also facing the same issue as I have annotated the data properly using the CVAT after training it on resnet 10 and 18 I found the images that are processed by the engine formed by int8 conversion crops improper objects and I have tried tweaking the model parameters and the pruning threshold too but nothing works and the bounding boxes are not forming the way I annotated . I am using the same setup as above

Please provide complete information as applicable to your setup.

• Hardware Platform (Jetson / GPU)
• DeepStream Version
• JetPack Version (valid for Jetson only)
• TensorRT Version
• NVIDIA GPU Driver Version (valid for GPU only)
• Issue Type( questions, new requirements, bugs)
• How to reproduce the issue ? (This is for bugs. Including which sample app is using, the configuration files content, the command line used and other details for reproducing)
• Requirement details( This is for new requirement. Including the module name-for which plugin or for which sample application, the function description)

Hardware Platform (Jetson / GPU) :GPU
**• DeepStream Version: 5.1
• NVIDIA GPU Driver Version (valid for GPU only): 11.2

I am expecting the image of the logs should be the same as in the bounding box as I am performing ANPR, I need to capture the number plate but what I am getting is cropped number plate.

Can you share some picture to show your issue? Thanks.


Please check.
One more problem I am facing is, when there is a high speeding vehicle, crossing the FOV, I am not able to capture the number plates.

What is the expect result? Do you mean BBox don’t cover all the plate number?

Expect result is I want to implement ANPR. Since the number plate, I am getting, is not proper/full, it is of no use.
Coming to BB, yes it does cover the full number plate.
Also, the bounding boxes are not appropriate for fast-moving objects.

Do you mean is the issue related with object moving speed?

yes, this issue is with moving objects.

In fact, in the high speeding vehicle, the percentage of capturing number plates is approx 20-30% only.

But the BBox is returned by model. Is it possible a model issue?

I don’t think this is a model issue. In the case of a high speeding vehicle, BBox does not form and thus does not return any number plate. In case, when there is a low speeding vehicle BBox is returned but the problem is the same as discussed above.