Yolov3 deepstream vs pytorch pipeline differences

Hello,

I have some question about running yolov3 in deepstream 6.0.1.

I have used yolov3 from darknet: YOLO: Real-Time Object Detection , trained it on custom dataset (GitHub - ultralytics/yolov3: YOLOv3 in PyTorch > ONNX > CoreML > TFLite) and deployed.
Everything works. Almost ^^

When I was training the model I have tested the model with some batch of pictures.
I got +/- 98% mAP on the dataset.

I have used the same dataset in my deepstream pipeline and I see some differences.

Some bboxes are just a little moved, some are not drawn at all.

I was trying to change some yolov3_config values (net-scale-factor, nms-iou-threshold etc) what is passed to nvinfer (config-file-path) but still the final mAP is much worse.

And maybe it’s important:
For instance for some picture, where I have 3 objects, confidence for all of them is above 60% (pytorch pipeline)
However, when I use deepstream pipeline (so the model is converted via trt) - the confidence is much lower (like 20-30%).

So maybe do you know what could be wrong here?

Of course I can append some of config files etc.

Thank you for help

Yes, please append your config file.
Besides please provide complete information as applicable to your setup, thanks.
• 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)
• The pipeline being used

• Hardware Platform (Jetson / GPU): GPU A2000
• DeepStream Version 6.0.1
• TensorRT Version 8.0.1-1+cuda11.3
• NVIDIA GPU Driver Version (valid for GPU only) 470.86
• The pipeline being used:

yolo configs:
yolov3-1cls.cfg (8.1 KB)
yolov3_config.txt (895 Bytes)

Hi @BonTo , Could you try to use yolov3 model in our demo to test the mAP? About how to get the model, you can refer the link below. Thanks
https://github.com/NVIDIA-AI-IOT/deepstream_tao_apps

There is no update from you for a period, assuming this is not an issue anymore.
Hence we are closing this topic. If need further support, please open a new one.
Thanks