Failed to parse ETLT file

Accelerated Computing Intelligent Video Analytics TAO Toolkit
1

Hello, I’m new with Isaac SDK 2020.1.
I was trying to train a custom object detection model with DetectNetv2 following the documentation. I can succesfully follow all the steps and generate the .etlt model. When I try to run inference on Isaac Sim Unity3D i get the following error passing the exported .etlt model

`2020-07-13 11:15:30.675 ERROR packages/ml/TensorRTInference.cpp@168: TRT ERROR: UffParser: Could not open /tmp/tmpfileCESgTX
2020-07-13 11:15:30.675 ERROR engine/alice/components/Codelet.cpp@229: Component ‘detect_net_inference.tensor_r_t_inference/isaac.ml.TensorRTInference’ of type ‘isaac::ml::TensorRTInference’ reported FAILURE:

Failed to parse ETLT file at /home/manuel/Desktop/tirocinio/models/resnet18_detector_cone.etlt

2020-07-13 11:15:30.675 ERROR engine/alice/backend/event_manager.cpp@42: Stopping node ‘detect_net_inference.tensor_r_t_inference’ because it reached status ‘FAILURE’
2020-07-13 11:15:30.675 ERROR engine/alice/components/Codelet.cpp@229: Component ‘detect_net_inference.tensor_r_t_inference/isaac.ml.TensorRTInference’ of type ‘isaac::ml::TensorRTInference’ reported FAILURE:

Could not read from plan (path is set to '/home/manuel/Desktop/tirocinio/models/resnet18_detector_cone.plan').failed to build the plan from frozen model /home/manuel/Desktop/tirocinio/models/resnet18_detector_cone.etlt

2020-07-13 11:15:30.675 ERROR engine/alice/backend/event_manager.cpp@42: Stopping node ‘detect_net_inference.tensor_r_t_inference’ because it reached status ‘FAILURE’`

Any suggestions? Thanks in advance

3

Could you please share the detailed steps for how to run inference on Isaac Sim Unity3D?

4

Ok, I solved manually setting the key environment variable at each step. Everything works fine now, thanks for you work!

6

Hi @r34ll0c please guide me how did you manage to solve

"
2020-07-13 11:15:30.675 ERROR packages/ml/TensorRTInference.cpp@168: TRT ERROR: UffParser: Could not open /tmp/tmpfileCESgTX
2020-07-13 11:15:30.675 ERROR engine/alice/components/Codelet.cpp@229: Component ‘detect_net_inference.tensor_r_t_inference/isaac.ml.TensorRTInference’ of type ‘isaac::ml::TensorRTInference’ reported FAILURE

"

it would be kind of you.

Regards

7

@r34ll0c What exactly did you do to come to this solution? I have just started working with custom object detection and I came to the same error where it Failed to parse ETLT file.

8

Please search keyword in TLT forum. If you still cannot find the solution, please create a new forum topic.

9

I figured out why I was getting this error. I had to update the “etlt_password” in the custom config file. I never changed to the one I had in the jupyter notebook.

{
  "config": {
    "detect_net_inference.tensor_r_t_inference": {
      "isaac.ml.TensorRTInference": {
        "model_file_path": "/home/jon/Dev/tlt-experiments/detectnet_v2/experiment_dir_final/resnet18_detector.etlt",
        "etlt_password": "tlt_encode", <-- CHANGE PASSWORD
        "max_batch_size": 1,
        "inference_mode": "Float32",
        "force_engine_update": true
      }
    },
    "detect_net_inference.detection_decoder": {
      ...
    }
  }
}