TDA4VM: Param.file generation error using tidl repo

Ramaseshan Subramanian

Part Number: TDA4VM

I have compiled my custom model yolov5s6 model with tidl repo, I could see most of the artifacts file has been generated but while creating param file I got this error.

------------------ Network Compiler Traces -----------------------------
successful Memory allocation
INFORMATION: [TIDL_ResizeLayer] Resize_107 Any resize ratio which is power of 2 and greater than 4 will be placed by combination of 4x4 resize layer and 2x2 resize layer. For example a 8x8 resize will be replaced by 4x4 resize followed by 2x2 resize.
INFORMATION: [TIDL_ResizeLayer] Resize_123 Any resize ratio which is power of 2 and greater than 4 will be placed by combination of 4x4 resize layer and 2x2 resize layer. For example a 8x8 resize will be replaced by 4x4 resize followed by 2x2 resize.
INFORMATION: [TIDL_ResizeLayer] Resize_139 Any resize ratio which is power of 2 and greater than 4 will be placed by combination of 4x4 resize layer and 2x2 resize layer. For example a 8x8 resize will be replaced by 4x4 resize followed by 2x2 resize.
****************************************************
**          3 WARNINGS          0 ERRORS          **
****************************************************
In TIDL_runtimesPostProcessNet 4
Process Process-1:
Traceback (most recent call last):
  File "/usr/lib/python3.10/multiprocessing/process.py", line 314, in _bootstrap
    self.run()
  File "/usr/lib/python3.10/multiprocessing/process.py", line 108, in run
    self._target(*self._args, **self._kwargs)
  File "/home/mugundan/edgeai-tidl-tools/examples/osrt_python/ort/onnxrt_ep.py", line 245, in run_model
    gen_param_yaml(delegate_options['artifacts_folder'], config, int(height), int(width))
  File "/home/mugundan/edgeai-tidl-tools/examples/osrt_python/common_utils.py", line 143, in gen_param_yaml
    dict_file[0]['postprocess']['detection_thr'] = 0.3
KeyError: 'postprocess'
************ in TIDL_subgraphRtDelete ************ 
 TIDL_deactivate is called with handle : 9b7e2000 
PREEMPTION: Removing priroty object with handle = 0x7fe59b7e2000 and targetPriority = 2,      Number of obejcts left are = 0, removed object with base  = 0x7fe59c22f000 and size =128
MEM: Deinit ... !!!
MEM: Alloc's: 26 alloc's of 336192635 bytes 
MEM: Free's : 26 free's  of 336192635 bytes 
MEM: Open's : 0 allocs  of 0 bytes 
MEM: Deinit ... Done !!!

over 1 year ago

0 Pratik Kedar over 1 year ago

TI__Mastermind 24041 points

Hi,

Can you help us with few details.

Is this issue is coming for above specified model ?

Can you confirm for std supported models param file generation is working at your host machine set up ?

What sdk tag you are using for edgai tidl tools ?

0 Ramaseshan Subramanian over 1 year ago in reply to Pratik Kedar

Intellectual 490 points

I could see that some of the tidl model param file has been generated successfully, I am using the newer version sdk 9.0 for compilation, I have compiled the model successfully and the only error I faced while generating param file.

 dict_file.append( {'last' : {
        'model_path' : os.path.join('/home/mugu/edgeai-tidl-tools/models/public/last.onnx'),
        'mean': [0, 0, 0],
        'std' : [0.003921568627,0.003921568627,0.003921568627],
        'num_images' : numImages,
        'num_classes': 9,
        'model_type': 'od',
        'meta_layers_names_list' : os.path.join('/home/mugu/edgeai-tidl-tools/models/public/last.prototxt'),
        'session_name' : 'onnxrt' ,
        'meta_arch_type' : 6,
        'target_device': 'pc'
    }})
    
    if(config['model_type'] == "od"):
        if(config['od_type'] == "SSD"):
            dict_file[0]['postprocess']['formatter'] = {'name' : 'DetectionBoxSL2BoxLS', 'src_indices' : [5,4]}
        elif(config['od_type'] == "HasDetectionPostProcLayer"):
            dict_file[0]['postprocess']['formatter'] = {'name' : 'DetectionYXYX2XYXY','src_indices' : [1,0,3,2]}
        
        dict_file[0]['postprocess']['detection_thr'] = 0.3

    with open(os.path.join(artifacts_folder_path, "param.yaml"), 'w') as file:
        documents = yaml.dump(dict_file[0], file)

    if (config['session_name'] == 'tflitert') or (config['session_name'] == 'onnxrt'):
        shutil.copy(config['model_path'], os.path.join(artifacts_folder_path,model_file_name))

0 Pratik Kedar over 1 year ago in reply to Ramaseshan Subramanian

TI__Mastermind 24041 points

Can you help me with exact sdk tag you are using ?

Ramaseshan Subramanian said:
I could see that some of the tidl model param file has been generated successfully, I

Can you list out the names of those models ?

For your current model (yolov5s6) is what you are facing issue with while generating the param.yaml file is my understanding correct ?

0 Ramaseshan Subramanian over 1 year ago in reply to Pratik Kedar

Intellectual 490 points

Hi Pratik thanks for the reply!

I have fixed this issue, the post process does not defined correctly and i could see that now param has been generated

0 Pratik Kedar over 1 year ago in reply to Ramaseshan Subramanian

TI__Mastermind 24041 points

Thats great to hear !

Can we close this issue then ?

0 Ramaseshan Subramanian over 1 year ago in reply to Pratik Kedar

Intellectual 490 points

Thanks Pratik

I have one last question, can we use the compiled artifacts using tidl repo for making inference or deployment on TDA4VM, is there any procedure or document to making deployment of compiled model on TDA4VM