I'm a hobby programmer. 
Intel Software Innovator member list
PINTO0309
Follow
Highlights
- Pro
Pinned
5,027 contributions in the last year
We're celebrating 100 million developers!
Activity overview
Contributed to
PINTO0309/onnx2tf,
PINTO0309/PINTO_model_zoo,
PINTO0309/simple-onnx-processing-tools
and 145 other
repositories
Contribution activity
January 2023
Created 272 commits in 11 repositories
Created 1 repository
- PINTO0309/depthai_simple Python
Opened 49 pull requests in 3 repositories
PINTO0309/onnx2tf
46
merged
1
closed
- Fixed NMS error when outputting to Keras (.h5)
-
Fixed Abort problem on Keras (.h5) output when using
np.dtype - Fixed to generate a tflite file with the same name as the input onnx file
-
Corrected the decision condition to enable
DepthwiseConv2D - Improved full validation algorithm
- Improved error messages
-
Fix
np.ndarrayof weights to be converted to tensor withtf.convert_to_tensor -
Reduces the frequency of
MatMulshape unmatch errors -
Improve processing of
Gatherto eliminatetf.where -
Tensors deleted from the TensorFlow model structure during the model optimization process are not comparable, so the status is rewritten to
Skipped -
[Experimental] Avoid
MUL requires one tensor that not less than second in all dimensions. -
Add
--optimization_for_gpu_delegateoption - [YOLOv8 FP16] TensorFlow's Resize operation casts to Float32 on its own, so we have to change it back to the original type.
- Remove unnecessary processing
-
Padding in
SYMMETRICmode will cause an error if the padding size is larger than the input size, so replaceCONSTANTwithSYMMETRIC -
disable_unnecessary_transpose, Skip special processing if the operation does not result in an error even if special processing is not performed -
[Experimental] Fixed a problem in which the transposition flag was wrongly determined when the shapes of all parts of the tensor except the batch size after
MatMulhad the same numerical value -
Added indication of maximum absolute error in case of
Unmatched -
Added option to compress
FlexTransposeto any number of dimensions - AveragePool and MaxPool padding handling bug fixes
-
Added
Skippeddecision for 1D tensor -
Exclude
ConstantandConstantOfShapefrom validation - Allow selection of whether to use normalized or denormalized data during validation
- [Experimental] Improved consistency check by downloading MS-COCO dummy data when model inputs appear to assume images
- Added ability to output only model weights in hdf5 format
- Some pull requests not shown.
PINTO0309/sam4onnx
1
merged
PINTO0309/sne4onnx
1
merged
Created an issue in PINTO0309/onnx2tf that received 13 comments
[FastestDet] The absolute error of AveragePool, GlobalAveragePool is quite large
Issue Type Others onnx2tf version number 1.5.12 onnx version number 1.13.0 tensorflow version number 2.10.0 Download URL for ONNX https://github.co…
13
comments
Opened 9 other issues in 2 repositories
PINTO0309/onnx2tf
7
closed
1
open
- [YOLOX-X] [Faster-RCNN] [YOLOv7] TensorFlow aborts when exporting a model with NMS to Keras (.h5)
-
[YOLOX-X] [Faster-RCNN] Error when outputting to h5 file.
AttributeError: 'numpy.dtype[int64]' object has no attribute 'item' - Implementation of strict mode
-
[MobileBERT] Add
--optimization_for_gpu_delegateoption - [MobileBERT] RecursionError: maximum recursion depth exceeded
-
[human_segmentation_pphumanseg_2021oct]
-cotofoption is used, Validation of 1D tensor always results inUnmatched -
[human_segmentation_pphumanseg_2021oct]
-cotofoption is used, ValueError: Output tensors of a Functional model must be the output of a TensorFlow Layer -
[YOLOX] Abort onnxruntime when
keepdims=FalseinReduceXXand the final tensor dimension is reduced to zero
Tarak200/XY-Single-Image-Deblurring
1
open
43
contributions
in private repositories
Jan 5 – Jan 27