PINTO0309
commented
1 year ago First, this problem requires an understanding of two issues related to the use of TFLite converters and the story of how onnx2tf works.
Your model appears to have 3 channels of input and 64 channels of non-image data, as shown in the image.
The tool will attempt to quantize using the default data set I have prepared if the user does not give data for calibration. The default data set refers to the 3-channel MS-COCO image data set.
https://github.com/PINTO0309/onnx2tf#cli-parameter
-qcind INPUT_NAME NUMPY_FILE_PATH MEAN STD, \
--quant_calib_input_op_name_np_data_path INPUT_NAME NUMPY_FILE_PATH MEAN STD
INPUT Name of OP and path of calibration data file (Numpy) for quantization and mean and std.
The specification can be omitted only when the input OP is a single 4D tensor image data.
If omitted, it is automatically calibrated using 20 normalized MS-COCO images.
The type of the input OP must be Float32.
Data for calibration must be pre-normalized to a range of 0 to 1.
-qcind {input_op_name} {numpy_file_path} {mean} {std}
Numpy file paths must be specified the same number of times as the number of input OPs.
Normalize the value of the input OP based on the tensor specified in mean and std.
(input_value - mean) / std
Tensors in Numpy file format must be in dimension order after conversion to TF.
Note that this is intended for deployment on low-resource devices,
so the batch size is limited to 1 only.
e.g.
The example below shows a case where there are three input OPs.
Assume input0 is 128x128 RGB image data.
In addition, input0 should be a value that has been divided by 255
in the preprocessing and normalized to a range between 0 and 1.
input1 and input2 assume the input of something that is not an image.
Because input1 and input2 assume something that is not an image,
the divisor is not 255 when normalizing from 0 to 1.
"n" is the number of calibration data.
ONNX INPUT shapes:
input0: [n,3,128,128]
mean: [1,3,1,1] -> [[[[0.485]],[[0.456]],[[0.406]]]]
std : [1,3,1,1] -> [[[[0.229]],[[0.224]],[[0.225]]]]
input1: [n,64,64]
mean: [1,64] -> [[0.1, ..., 0.64]]
std : [1,64] -> [[0.05, ..., 0.08]]
input2: [n,5]
mean: [1] -> [0.3]
std : [1] -> [0.07]
TensorFlow INPUT shapes (Numpy file ndarray shapes):
input0: [n,128,128,3]
mean: [1,1,1,3] -> [[[[0.485, 0.456, 0.406]]]]
std : [1,1,1,3] -> [[[[0.229, 0.224, 0.225]]]]
input1: [n,64,64]
mean: [1,64] -> [[0.1, ..., 0.64]]
std : [1,64] -> [[0.05, ..., 0.08]]
input2: [n,5]
mean: [1] -> [0.3]
std : [1] -> [0.07]
-qcind "input0" "../input0.npy" [[[[0.485, 0.456, 0.406]]]] [[[[0.229, 0.224, 0.225]]]]
-qcind "input1" "./input1.npy" [[0.1, ..., 0.64]] [[0.05, ..., 0.08]]
-qcind "input2" "input2.npy" [0.3] [0.07]Debugging allows you to see the shape of the data set for calibration. It is shown in the figure below.
This means that if you need to use special input data other than images, you will need to pass the calibration data set to TFLite Converter yourself. The 64-channel data is too special, and it is not possible to determine just from the structure of the model what kind of calibration should be done for onnx2tf in the first place. Therefore, you need to generate your own Numpy dataset and pass it to onnx2tf, as described in the README tutorial transcribed above.
Also, to give you an idea of a potential problem that may arise a little further down the road in exchanging this topic, it is possible that TFLite Converter may not be able to successfully complete quantization for models with more than two inputs. This is not a problem with onnx2tf, but rather with the specification of the TFLite Converter itself.
github-actions[bot]
PengWei-Lin
Issue Type
Others
onnx2tf version number
1.7.6
onnx version number
1.13.1
tensorflow version number
2.12.0rc0
Download URL for ONNX
https://drive.google.com/file/d/1yKCN8_2ayeBLOhd6bgeL55pu_HXClO1L/view?usp=share_link
Parameter Replacement JSON
Description