TFLite Conversion error

VIM4
1

Hello I’m trying to convert a tflite model to use on NPU stack, and I’m facing some issues, here is the command line used by ‘convert-adla.sh’:

$adla_convert --model-type tflite \
        --model /home/gustavo/sources/coffee_mobilenet/saved_models/tflite/mobilenet.tflite \
        --dtypes float32 \
        --inputs input_layer_1 --input-shapes 224,224,3 \
        --quantize-dtype int8 --outdir tflite_output \
        --channel-mean-value "0,0,0,256" \
        --source-file dataset.txt \
        --target-platform PRODUCT_PID0XA001
when I start the conversion proccess, these messages are shown:
DOCKER_RUN:docker run -it --name npu-vim4 --rm -v /home/gustavo/vim4_npu_sdk:/home/khadas/npu -v /etc/localtime:/etc/localtime:ro -v /etc/timezone:/etc/timezone:ro -v /home/gustavo:/home/gustavo numbqq/npu-vim4
I Namespace(batch_size=1, channel_mean_value='0,0,0,256', default_ranges_max=None, default_ranges_min=None, disable_per_channel='False', dtypes='float32', inference_input_type=None, inference_output_type=None, input_shapes='224,224,3', inputs='input_layer_1', iterations=1, mean=0, model='/home/gustavo/sources/coffee_mobilenet/saved_models/tflite/mobilenet.tflite', model_type='tflite', outdir='tflite_output', outputs=None, quantize_dtype='int8', reorder_channel=None, shape_with_batch=None, source_file='dataset.txt', std_dev=1, target_platform='PRODUCT_PID0XA001', weights=None)
I Adla_tool: step_0 enter, begin to load original model.
I Adla_tool: step_0 out, load original model successfully.
I Adla_tool: step_1 enter, frontend begin to parse model.
I Adla_tool: step_1 out, frontend parse model successfully.
I Adla_tool version:v1.2.0,2022.10.28
I Adla_tool: step_2 enter, begin to convert model to ir.
I Input_config input_name:input_layer_1, input_shape:(224, 224, 3), batch:1,dtype:float32
tvm/runtime/ndarray.py:197: DeprecationWarning: NDArray.asnumpy() will be deprecated in TVM v0.8 release. Please use NDArray.numpy() instead.
I Process add_layer[1] @nn.conv2d
I Parse layer @Conv...
E 'serving_default_input_layer_1:0'
I ----------------Warning(0)----------------

I’m a bit confused, I’m not understanding what is the erro, if someone pelase help me to understand I’ll be gratefull.

Thanks and best regards,
Gustavo

2

Hello @Gustavo_Ferreira_Pal ,

Could you provide your model?

3

HI @Louis-Cheng-Liu,

sure here it is link to tflite model

4

Hello @Gustavo_Ferreira_Pal ,

I have get your model and will notify you at once when we find the solution.

5

Thanks for all your efforts @Louis-Cheng-Liu, I appreciate your help

6

Hello @Gustavo_Ferreira_Pal ,

inputs need to use model input.name.

image
image736×180 5.38 KB

$adla_convert --model-type tflite \
        --model /home/gustavo/sources/coffee_mobilenet/saved_models/tflite/mobilenet.tflite \
        --dtypes float32 \
        --inputs serving_default_input_layer_1:0 --input-shapes 224,224,3 \
        --quantize-dtype int8 --outdir tflite_output \
        --channel-mean-value "0,0,0,256" \
        --source-file dataset.txt \
        --target-platform PRODUCT_PID0XA001
7

Hi @Louis-Cheng-Liu, thank you for your help, the conversion is completed successfully.