[DeepStream 7.1] ResNet18 Fall Classifier ONNX Model Not Inferencing (Previously Worked in DS 7.0)

[sagar0]

I’m facing issues running a ResNet18-based fall classifier model on DeepStream 7.1. This model was originally trained using the NVIDIA TAO Toolkit and exported to ONNX. It worked perfectly on DeepStream 7.0, but after migrating to DS 7.1, the same ONNX model does not infer anymore — no output or classification is generated.

Details:

• DeepStream version: 7.1
• Model framework: ResNet18, trained via TAO Toolkit
• Export format: ONNX, exported using TAO export command (used successfully in DS 7.0)
• Model precision: FP32
• DeepStream config: Model is used via nvinfer plugin with onnx-file specified and model-engine-file path set.

inception

[Fiona.Chen]

The only change is the TensorRT version. DeepStream 7.0 is TensorRT 8.6.1, DeepStream 7.1 is TensorRT 10.3.

Please check with TAO toolkit whether your ONNX model work in TensorRT 10.3