PeopleNet not Detecting Bags

Accelerated Computing Intelligent Video Analytics TAO Toolkit
1

Hi,

I tested peoplenet on pruned_quantized_decrypted_v2.3.4 and deployable_quantized_onnx_v2.6.3 and realized that the model detects person and face well but detects no bag. Appreciate if you could point out what went wrong.

Below are the info for my implementation:

  1. preprocessing step from tao-toolkit-triton-apps/tao_triton/python/model/detectnet_model.py at main · NVIDIA-AI-IOT/tao-toolkit-triton-apps · GitHub
  2. postprocessing config from tao-toolkit-triton-apps/tao_triton/python/clustering_specs/clustering_config_peoplenet.prototxt at main · NVIDIA-AI-IOT/tao-toolkit-triton-apps · GitHub
  3. postprocessing steps from tao-toolkit-triton-apps/tao_triton/python/postprocessing/detectnet_processor.py at main · NVIDIA-AI-IOT/tao-toolkit-triton-apps · GitHub

Sample Result

image
image1920×1080 158 KB

2

Please refer to Bag detection not working for TAO Peoplenet model v2.6 - #2 by Morganh.

3

Hi,

I downloaded resnet34_peoplenet_pruned.etlt (pruned_v2.0) and attempted converting .etlt to .onnx using the command below

!tao model detectnet_v2 export \
                  -m $USER_EXPERIMENT_DIR/model.etlt \
                  -k tlt_encode \
                  -o $USER_EXPERIMENT_DIR/model1.onnx

The terminal output is as follows:

image
image1852×718 82.4 KB

Despite getting a ‘passed’, no model is being saved out. Appreciate your advice on this.

Thank you!

4

Please download .tlt file and export to onnx.