%90 val_accuracy with TLT but no bounding boxes with Deepstream Python api

[Python]

I have trained detectnet_v2 resnet34 model and evaluated with testing dataset. The MAp is %90, however when i deploy it and inference with jetson Tx2 module, there is no bounding boxes, i want to see at least false positives but nothings seems, fps is 25 and its the max fps that my camera provides.

I double checked every possible problem but could not find, any possible idea to help will be welcomed.

• Hardware Platform (Jetson / GPU) Jetson tx2
• DeepStream Version 5.
• JetPack Version (valid for Jetson only) 4.4
• TensorRT Version 7.1
• NVIDIA GPU Driver Version (valid for GPU only)

[bcao]

Seems it’s related to TLT, coudl you create a topic in TLT forum and make sure the mode inference result is correct without DS pipeline.