While converting onnx to tflite TILE layer is getting ignored. about onnx2tf

[PINTO0309]

What is it you are trying to do when a structurally perfectly correct model is being generated?

Your model Tile is garbage. I recommend that you look into the process called "broadcast".

onnx2tf -i sample.onnx -cotof

Model conversion started ============================================================
INFO: input_op_name: input_1 shape: [1, 4, 56, 56] dtype: float32
INFO: input_op_name: input_2 shape: [1, 96, 1, 1] dtype: float32

INFO: 3 / 5
INFO: onnx_op_type: Slice onnx_op_name: sng_Slice_0
INFO:  input_name.1: input_1 shape: [1, 4, 56, 56] dtype: float32
INFO:  input_name.2: slice0 shape: [1] dtype: int64
INFO:  input_name.3: slice1 shape: [1] dtype: int64
INFO:  input_name.4: slice2 shape: [1] dtype: int64
INFO:  input_name.5: slice2 shape: [1] dtype: int64
INFO:  output_name.1: slice shape: [1, 1, 56, 56] dtype: float32
INFO: tf_op_type: strided_slice
INFO:  input.1.input_: name: input_1 shape: (1, 56, 56, 4) dtype: <dtype: 'float32'> 
INFO:  input.2.begin: shape: (1,) dtype: int64 
INFO:  input.3.end: shape: (1,) dtype: int64 
INFO:  input.4.strides: shape: (1,) dtype: <dtype: 'int64'> 
INFO:  output.1.output: name: tf.strided_slice/StridedSlice:0 shape: (1, 56, 56, 1) dtype: <dtype: 'float32'> 

INFO: 4 / 5
INFO: onnx_op_type: Tile onnx_op_name: sng_Tile_1
INFO:  input_name.1: slice shape: [1, 1, 56, 56] dtype: float32
INFO:  input_name.2: tile1 shape: [4] dtype: int64
INFO:  output_name.1: tile shape: [1, 96, 56, 56] dtype: float32
INFO: tf_op_type: tile
INFO:  input.1.input: name: tf.strided_slice/StridedSlice:0 shape: (1, 56, 56, 1) dtype: <dtype: 'float32'> 
INFO:  input.2.multiples: shape: (4,) dtype: int64 
INFO:  output.1.output: name: tf.tile_1/Tile:0 shape: (1, 56, 56, 96) dtype: <dtype: 'float32'> 

INFO: 5 / 5
INFO: onnx_op_type: Mul onnx_op_name: sng_Mul_2
INFO:  input_name.1: tile shape: [1, 96, 56, 56] dtype: float32
INFO:  input_name.2: input_2 shape: [1, 96, 1, 1] dtype: float32
INFO:  output_name.1: output shape: [1, 96, 56, 56] dtype: float32
INFO: tf_op_type: multiply
INFO:  input.1.x: name: tf.tile_1/Tile:0 shape: (1, 56, 56, 96) dtype: <dtype: 'float32'> 
INFO:  input.2.y: name: input_2 shape: (1, 1, 1, 96) dtype: <dtype: 'float32'> 
INFO:  output.1.output: name: tf.math.multiply_3/Mul:0 shape: (1, 56, 56, 96) dtype: <dtype: 'float32'> 

saved_model output started ==========================================================
saved_model output complete!
Float32 tflite output complete!
Float16 tflite output complete!

ONNX and TF output value validation started =========================================
INFO: validation_conditions: np.allclose(onnx_outputs, tf_outputs, rtol=0.0, atol=0.0001, equal_nan=True)
INFO: onnx_output_name: slice tf_output_name: tf.strided_slice/StridedSlice:0 shape: (1, 1, 56, 56) dtype: float32 validate_result:  Matches 
INFO: onnx_output_name: tile tf_output_name: tf.tile_1/Tile:0 shape: (1, 96, 56, 56) dtype: float32 validate_result:  Matches 
INFO: onnx_output_name: output tf_output_name: tf.math.multiply_3/Mul:0 shape: (1, 96, 56, 56) dtype: float32 validate_result:  Matches 

[rjharahul21]

I am getting accuracy drop while running model on gpu.