MMdnn is a set of tools to help users inter-operate among different deep learning frameworks. E.g. model conversion and visualization. Convert models between Caffe, Keras, MXNet, Tensorflow, CNTK, PyTorch Onnx and CoreML.

Last update: Jan 9, 2023

Overview

MMdnn

MMdnn is a comprehensive and cross-framework tool to convert, visualize and diagnose deep learning (DL) models. The "MM" stands for model management, and "dnn" is the acronym of deep neural network.

Major features include:

Model Conversion
- We implement a universal converter to convert DL models between frameworks, which means you can train a model with one framework and deploy it with another.
Model Retraining
- During the model conversion, we generate some code snippets to simplify later retraining or inference.
Model Search & Visualization
- We provide a model collection to help you find some popular models.
- We provide a model visualizer to display the network architecture more intuitively.
Model Deployment
- We provide some guidelines to help you deploy DL models to another hardware platform.
  - Android
  - Serving
- We provide a guide to help you accelerate inference with TensorRT.
  - TensorRT

Related Projects

Targeting at openness and advancing state-of-art technology, Microsoft Research (MSR) and Microsoft Software Technology Center (STC) had also released few other open source projects:

OpenPAI : an open source platform that provides complete AI model training and resource management capabilities, it is easy to extend and supports on-premise, cloud and hybrid environments in various scale.
FrameworkController : an open source general-purpose Kubernetes Pod Controller that orchestrate all kinds of applications on Kubernetes by a single controller.
NNI : a lightweight but powerful toolkit to help users automate Feature Engineering, Neural Architecture Search, Hyperparameter Tuning and Model Compression.
NeuronBlocks : an NLP deep learning modeling toolkit that helps engineers to build DNN models like playing Lego. The main goal of this toolkit is to minimize developing cost for NLP deep neural network model building, including both training and inference stages.
SPTAG : Space Partition Tree And Graph (SPTAG) is an open source library for large scale vector approximate nearest neighbor search scenario.

We encourage researchers, developers and students to leverage these projects to boost their AI / Deep Learning productivity.

Installation

Install manually

You can get a stable version of MMdnn by

pip install mmdnn

And make sure to have Python installed or you can try the newest version by

pip install -U git+https://github.com/Microsoft/MMdnn.git@master

Install with docker image

MMdnn provides a docker image, which packages MMdnn and Deep Learning frameworks that we support as well as other dependencies. You can easily try the image with the following steps:

Install Docker Community Edition(CE)

Learn more about how to install docker
Pull MMdnn docker image
```
docker pull mmdnn/mmdnn:cpu.small
```
Run image in an interactive mode
```
docker run -it mmdnn/mmdnn:cpu.small
```

Features

Model Conversion

Across the industry and academia, there are a number of existing frameworks available for developers and researchers to design a model, where each framework has its own network structure definition and saving model format. The gaps between frameworks impede the inter-operation of the models.

We provide a model converter to help developers convert models between frameworks through an intermediate representation format.

Support frameworks

[Note] You can click the links to get detailed README of each framework.

Caffe
Microsoft Cognitive Toolkit (CNTK)
CoreML
Keras
MXNet
ONNX (Destination only)
PyTorch
TensorFlow (Experimental) (We highly recommend you read the README of TensorFlow first)
DarkNet (Source only, Experiment)

Tested models

The model conversion between currently supported frameworks is tested on some ImageNet models.

Models	Caffe	Keras	TensorFlow	CNTK	MXNet	PyTorch	CoreML	ONNX
VGG 19	√	√	√	√	√	√	√	√
Inception V1	√	√	√	√	√	√	√	√
Inception V3	√	√	√	√	√	√	√	√
Inception V4	√	√	√	o	√	√	√	√
ResNet V1	×	√	√	o	√	√	√	√
ResNet V2	√	√	√	√	√	√	√	√
MobileNet V1	×	√	√	o	√	√	√	√
MobileNet V2	×	√	√	o	√	√	√	√
Xception	√	√	√	o	×	√	√	√
SqueezeNet	√	√	√	√	√	√	√	√
DenseNet	√	√	√	√	√	√	√	√
NASNet	x	√	√	o	√	√	√	x
ResNext	√	√	√	√	√	√	√	√
voc FCN			√	√
Yolo3		√		√

Usage

One command to achieve the conversion. Using TensorFlow ResNet V2 152 to PyTorch as our example.

$ mmdownload -f tensorflow -n resnet_v2_152 -o ./
$ mmconvert -sf tensorflow -in imagenet_resnet_v2_152.ckpt.meta -iw imagenet_resnet_v2_152.ckpt --dstNodeName MMdnn_Output -df pytorch -om tf_resnet_to_pth.pth

Done.

On-going frameworks

Torch7 (help wanted)
Chainer (help wanted)

On-going Models

Face Detection
Semantic Segmentation
Image Style Transfer
Object Detection
RNN

Model Visualization

We provide a local visualizer to display the network architecture of a deep learning model. Please refer to the instruction.

Examples

Official Tutorial

Users' Examples

Contributing

Most contributions require you to agree to a Contributor License Agreement (CLA) declaring that you have the right to and actually do, grant us the rights to use your contribution. For details, visit https://cla.microsoft.com.

When you submit a pull request, a CLA-bot will automatically determine whether you need to provide a CLA and decorate the PR appropriately (e.g., label, comment). Simply follow the instructions provided by the bot. You will only need to do this once across all repos using our CLA.

This project has adopted the Microsoft Open Source Code of Conduct. For more information see the Code of Conduct FAQ or contact [email protected] with any additional questions or comments.

Intermediate Representation

The intermediate representation stores the network architecture in protobuf binary and pre-trained weights in NumPy native format.

[Note!] Currently the IR weights data is in NHWC (channel last) format.

Details are in ops.txt and graph.proto. New operators and any comments are welcome.

Frameworks

We are working on other frameworks conversion and visualization, such as PyTorch, CoreML and so on. We're investigating more RNN related operators. Any contributions and suggestions are welcome! Details in Contribution Guideline.

Authors

Yu Liu (Peking University): Project Developer & Maintainer

Cheng CHEN (Microsoft Research Asia): Caffe, CNTK, CoreML Emitter, Keras, MXNet, TensorFlow

Jiahao YAO (Peking University): CoreML, MXNet Emitter, PyTorch Parser; HomePage

Ru ZHANG (Chinese Academy of Sciences): CoreML Emitter, DarkNet Parser, Keras, TensorFlow frozen graph Parser; Yolo and SSD models; Tests

Yuhao ZHOU (Shanghai Jiao Tong University): MXNet

Tingting QIN (Microsoft Research Asia): Caffe Emitter

Tong ZHAN (Microsoft): ONNX Emitter

Qianwen WANG (Hong Kong University of Science and Technology): Visualization

Acknowledgements

Thanks to Saumitro Dasgupta, the initial code of caffe -> IR converting is references to his project caffe-tensorflow.

License

Licensed under the MIT license.

Comments

Convert Model from MXNet to PyTorch

Dear @kitstar, Thank you for your nice repository. I have a pre-trained ResNet152 model on MXNet and I want to convert it to PyTorch. Would you please kindly guide me to do that?
question

opened by ahkarami 26
Cannot convert TF Mobilenet V2 to Caffe

Platform : Centos 7

Python version : 2.7

Source framework with version :Tensorflow 1.8.0

Destination framework with version : Caffe 1.0

I use the official Pre-trained model Mobilenet V1(http://download.tensorflow.org/models/mobilenet_v1_2018_02_22/mobilenet_v1_1.0_224.tgz) and success converted to Caffe model.

But I failed converting Pre-trained model Mobilenet V2(https://storage.googleapis.com/mobilenet_v2/checkpoints/mobilenet_v2_1.0_224.tgz).

Got error “Check failed : _data” after "MobilenetV2_Logits_Conv2d_1c_1x1_Conv2D -> MobilenetV2_Logits_Conv2d_1c_1x1_Conv2D"

Is TF Mobilenet V2 to Caffe not supportted?

opened by monckxqq 19
AttributeError: module 'keras.applications.mobilenet' has no attribute 'relu6'
Platform (like ubuntu 16.04/win10): macOS 10.13.5

Python version: 3.6

Source framework with version (like Tensorflow 1.4.1 with GPU): Tensorflow 1.8

Destination framework with version (like CNTK 2.3 with GPU): toIR

Pre-trained model path (webpath or webdisk path): ./version0056.h5

Running scripts: python -m mmdnn.conversion._script.convertToIR -f keras -d connect4_mobilenet -w version0056.h5

I have created a keras model using the example found here: https://github.com/AppliedDataSciencePartners/DeepReinforcementLearning

My ultimate goal is to try and import this into coreML to test with an iOS app. When I first try to convert to IR in order to convert to coreML I get the following issue:

Using TensorFlow backend. Traceback (most recent call last): File "/anaconda3/bin/mmconvert", line 11, in <module> sys.exit(_main()) File "/anaconda3/lib/python3.6/site-packages/mmdnn/conversion/_script/convert.py", line 102, in _main ret = convertToIR._convert(ir_args) File "/anaconda3/lib/python3.6/site-packages/mmdnn/conversion/_script/convertToIR.py", line 39, in _convert parser = Keras2Parser(model) File "/anaconda3/lib/python3.6/site-packages/mmdnn/conversion/keras/keras2_parser.py", line 94, in __init__ 'relu6': _keras.applications.mobilenet.relu6, AttributeError: module 'keras.applications.mobilenet' has no attribute 'relu6'

Seems like perhaps relu6 is not supported or something like that? Sorry new to this and trying to teach myself how this all fits together!
opened by pkl728 19
Convert ResNet101 from TensorFlow to PyTorch

Dear @kitstar, I want to convert a ResNet V1 101 model (from TF-Slim) to PyTorch. Would you please kindly help me to do that? Just as another suggestion, I think it would be great if you create a README.md file for PyTorch conversion section.

opened by ahkarami 19
How to Converting Mxnet to Tensorflow model in serving format.

Basically I have models which are already developed in Mxnet and I want to convert them into tensorflow serving format without performing retraining.

So for performing conversion from Mxnet to Tensorflow I have used this Intermediate layer created by Microsoft Mdnn.

Using this tool/library I was able to convert Mxnet model files to Tensorflow checkpoint format:

These are initial Mxnet model files. And this is my flow of conversion.

Mxnet_model => IR format => Conversion code => Checkpoint => Tensorflow Model, Serving format

IR = intermediate representation I have successfully converted till checkpoint files using this instruction Mxnet to IR thenIR to Tensorflow checkpoint

This final structure of tensorflow checkpoint file are:

The only step remaining is to convert and save these checkpoint files to .pb and variable folder format using this SaveBuilder function Saving using this function is essentials because its is the only valid format to save the model, inorder to serve the model on tensorflow serving

This is how the final structure of the converted model must look like in order to get served using TF-serving:

This is the accurate structure which TF-serving accept and generate predictions for.

I tried using these scripts freeze_graph.py and incepiton_save_model.py but nothing came out they have some arguments which I don't have files to pass.

Help!!! Is there a way? I am trying since last 3days but couldn't find anything. Thanks in advance.
question

opened by gr8Adakron 17
Error while converting from IR to CNTK (leaky_relu and upsampling2d)

Platform : win 10

Python version: 3.5.2

Source framework with version : keras with tensorflow 1.7.0 on CPU

Destination framework with version : CNTK 2.4 on CPU

Running scripts: 1. Convert the pre-trained model files to intermediate representation $ mmtoir -f keras -w yolo.h5 -o yolov3

2. Convert the IR files to CNTK models $ mmtocode -f cntk -d converted_cntk.py -n yolov3.pb -w yolov3.npy

Description: Converted trained model yolo.h5 from Keras -> IR. Got an error while converting from IR->CNTK. The error while running the second command is at link: https://pastebin.com/pxBVmj10

opened by almeida29 16
[Group convolution in Keras] ResNeXt mxnet -> IR -> keras

Hi Thank you for a great covert tool.

I am trying to convert from mxnet resnext to keras. symbol file: http://data.mxnet.io/models/imagenet/resnext/101-layers/resnext-101-64x4d-symbol.json param file: http://data.mxnet.io/models/imagenet/resnext/101-layers/resnext-101-64x4d-0000.params

I could convert from mxnet to IR with no error,

python -m mmdnn.conversion._script.convertToIR -f mxnet -n resnext-101-64x4d-symbol.json -w resnext-101-64x4d-0000.params -d resnext-101-64x4d --inputShape 3 224 224

but failed to convert from IR to keras with an error below. Would you support this model?

Regards,

python -m mmdnn.conversion._script.IRToCode -f keras --IRModelPath resnext-101-64x4d.pb --dstModelPath keras_resnext-101-64x4d.py

Parse file [resnext-101-64x4d.pb] with binary format successfully. Traceback (most recent call last): File "C:\Anaconda3\lib\runpy.py", line 193, in _run_module_as_main "main", mod_spec) File "C:\Anaconda3\lib\runpy.py", line 85, in _run_code exec(code, run_globals) File "C:\Anaconda3\lib\site-packages\mmdnn\conversion_script\IRToCode.py", line 120, in _main() File "C:\Anaconda3\lib\site-packages\mmdnn\conversion_script\IRToCode.py", line 115, in _main ret = _convert(args) File "C:\Anaconda3\lib\site-packages\mmdnn\conversion_script\IRToCode.py", line 56, in _convert emitter.run(args.dstModelPath, args.dstWeightPath, args.phase) File "C:\Anaconda3\lib\site-packages\mmdnn\conversion\common\DataStructure\emitter.py", line 21, in run self.save_code(dstNetworkPath, phase) File "C:\Anaconda3\lib\site-packages\mmdnn\conversion\common\DataStructure\emitter.py", line 53, in save_code code = self.gen_code(phase) File "C:\Anaconda3\lib\site-packages\mmdnn\conversion\keras\keras2_emitter.py", line 95, in gen_code func(current_node) File "C:\Anaconda3\lib\site-packages\mmdnn\conversion\keras\keras2_emitter.py", line 194, in emit_Conv return self._emit_convolution(IR_node, 'layers.Conv{}D'.format(dim)) File "C:\Anaconda3\lib\site-packages\mmdnn\conversion\keras\keras2_emitter.py", line 179, in _emit_convolution input_node, padding = self._defuse_padding(IR_node) File "C:\Anaconda3\lib\site-packages\mmdnn\conversion\keras\keras2_emitter.py", line 160, in _defuse_padding padding = self._convert_padding(padding) File "C:\Anaconda3\lib\site-packages\mmdnn\conversion\keras\keras2_emitter.py", line 139, in _convert_padding padding = convert_onnx_pad_to_tf(padding)[1:-1] File "C:\Anaconda3\lib\site-packages\mmdnn\conversion\common\utils.py", line 62, in convert_onnx_pad_to_tf return np.transpose(np.array(pads).reshape([2, -1])).reshape(-1, 2).tolist()

ValueError: cannot reshape array of size 1 into shape (2,newaxis)
bug enhancement help wanted

opened by kamikawa 14
CaffeEmitter has not supported operator [ResizeBilinear].
When I turn a tensorflow pose model into caffe model, it prompts the error "CaffeEmitter has not supported operator [ResizeBilinear].", my tensorflow pose model has this operator, So how can I solve it? I have two thoughts:

Firstly add the ResizeBilinear layer in cafffe, then I use mmdnn, Do mmdnn support newly add layers in caffe? What does mmdnn work?

Another thought is to convert tensorflow model to caffe model directly because ResizeBilinear layer doesn't have weights, so I can remove correlation ResizeBilinear layer in my model structure, then I add these layers in caffe model prototxt manually(Assuming I have implemented ResizeBilinear layer in caffe).
opened by ujsyehao 13
IR->Caffe?

Hi, In the Caffe directory there is no instruction for IR ->Caffe conversion. Is this conversion supported? I want to convert the weights in keras to caffe. How can I do this conversion using MMdnn? Thank you.

opened by anwesha94 13
Added PRelu to CoreML emitter.

Note: I haven't tested yet if this gives the correct output, but I am assuming it should.

Also, you should consider adding .DS_Store to the .gitignore, since these files are very prevalant when working on a mac.

opened by galli-leo 11

Cannot convert caffe model with "Reshape" layer

When the deploy.prototxt file contains layer of type "Reshape", convertToIR script crashes with error:

Traceback (most recent call last):
  File "/usr/lib/python2.7/runpy.py", line 174, in _run_module_as_main
    "__main__", fname, loader, pkg_name)
  File "/usr/lib/python2.7/runpy.py", line 72, in _run_code
    exec code in run_globals
  File "/usr/local/lib/python2.7/dist-packages/mmdnn/conversion/_script/convertToIR.py", line 159, in <module>
    _main()
  File "/usr/local/lib/python2.7/dist-packages/mmdnn/conversion/_script/convertToIR.py", line 154, in _main
    ret = _convert(args)
  File "/usr/local/lib/python2.7/dist-packages/mmdnn/conversion/_script/convertToIR.py", line 9, in _convert
    transformer = CaffeTransformer(args.network, args.weights, "tensorflow", args.inputShape, phase = args.caffePhase)
  File "/usr/local/lib/python2.7/dist-packages/mmdnn/conversion/caffe/transformer.py", line 308, in __init__
    graph = GraphBuilder(def_path, self.input_shape, self.is_train_proto, phase).build()
  File "/usr/local/lib/python2.7/dist-packages/mmdnn/conversion/caffe/graph.py", line 444, in build
    graph.compute_output_shapes(self.model)
  File "/usr/local/lib/python2.7/dist-packages/mmdnn/conversion/caffe/graph.py", line 265, in compute_output_shapes
    node.output_shape = TensorShape(*NodeKind.compute_output_shape(node))
  File "/usr/local/lib/python2.7/dist-packages/mmdnn/conversion/caffe/graph.py", line 125, in compute_output_shape
    return LAYER_DESCRIPTORS[node.kind](node)
KeyError: None

enhancement

opened by andrusza2 11

How can I convert tf(.pb) to pytorch?

Platform (like ubuntu 16.04/win10): ubuntu 20.04

Python version: 3.7

Source framework with version (like Tensorflow 1.4.1 with GPU): tf 1.14,

Destination framework with version (like CNTK 2.3 with GPU): torch 1.12.1

Pre-trained model path (webpath or webdisk path):

Running scripts:

mmconvert --srcFramework tensorflow --inputWeight saved_model.pb --inputNetwork saved_model.pb --dstFramework pytorch --outputModel tf2torch_saved_model

my error UnicodeDecodeError: 'utf-8' codec can't decode byte 0x9c in position 3: invalid start byte

opened by eric9687 0
Pytorch model GFPGANv1.3.pth

Platform Win 7 Python version: 3.8 Source framework with version Tensorflow 1.4.1 without GPU Pre-trained model path (webpath or webdisk path): Could you test this model ? GFPGANv1.3.pth

opened by magicse 0
Create CITATION.cff

This PR adds a CITATION.cff file, styled after the one used for TensorFlow.

The paper used for the DOI is the ACM paper for mmDNN, which mirrors how other deep learning software projects use this feature.

Why is the valuable? It improves discoverability of the ACM paper for GitHub users, and makes it more likely that the correct citation will be used when the system is used in research (rather than just citing the GitHub URL).

opened by Wheest 1

Getting ValueError: axes don't match array

Platform (like ubuntu 16.04/win10): Ubuntu 22.04 LTS

Python version: conda 4.13.0 Python 3.10.4

Source framework with version (like Tensorflow 1.4.1 with GPU): Caffe

Destination framework with version (like CNTK 2.3 with GPU): Tensorflow

Pre-trained model path (webpath or webdisk path): https://github.com/henzler/single-image-tomography

Running scripts: mmconvert -sf caffe -in deploy.prototxt -iw hourglass256x128x128_iter_150000.caffemodel -df tensorflow -om caffe

I want to run this model on my comptuer. I'm trying to convert it from Caffe to Tensorflow, but I'm having some difficulty in doing it. I really don't know what I'm doing or where I'm doing things wrong and I've been stuck here for hours. I would really be glad if someone could help.

I'm currently getting this error:

------------------------------------------------------------
    WARNING: PyCaffe not found!
    Falling back to a pure protocol buffer implementation.
    * Conversions will be drastically slower.
    * This backend is UNTESTED!
------------------------------------------------------------

Warning: parameters not reshaped for node: [BatchNorm] bn_conv1_b
Warning: parameters not reshaped for node: [BatchNorm] bn1_branch1
Warning: parameters not reshaped for node: [BatchNorm] bn1_branch2a
Warning: parameters not reshaped for node: [BatchNorm] bn1_branch2b
Warning: parameters not reshaped for node: [BatchNorm] bn1_branch2c
Warning: parameters not reshaped for node: [BatchNorm] bn4_branch2a
Warning: parameters not reshaped for node: [BatchNorm] bn4_branch2b
Warning: parameters not reshaped for node: [BatchNorm] bn4_branch2c
Warning: parameters not reshaped for node: [BatchNorm] bn5_branch2a
Warning: parameters not reshaped for node: [BatchNorm] bn5_branch2b
Warning: parameters not reshaped for node: [BatchNorm] bn5_branch2c
Warning: parameters not reshaped for node: [BatchNorm] bn6_branch1
Warning: parameters not reshaped for node: [BatchNorm] bn6_branch2a
Warning: parameters not reshaped for node: [BatchNorm] bn6_branch2b
Warning: parameters not reshaped for node: [BatchNorm] bn6_branch2c
Warning: parameters not reshaped for node: [BatchNorm] bnhg1_low1_branch2a
Warning: parameters not reshaped for node: [BatchNorm] bnhg1_low1_branch2b
Warning: parameters not reshaped for node: [BatchNorm] bnhg1_low1_branch2c
Warning: parameters not reshaped for node: [BatchNorm] bnhg1_low2_branch2a
Warning: parameters not reshaped for node: [BatchNorm] bnhg1_low2_branch2b
Warning: parameters not reshaped for node: [BatchNorm] bnhg1_low2_branch2c
Warning: parameters not reshaped for node: [BatchNorm] bnhg1_low5_branch2a
Warning: parameters not reshaped for node: [BatchNorm] bnhg1_low5_branch2b
Warning: parameters not reshaped for node: [BatchNorm] bnhg1_low5_branch2c
Warning: parameters not reshaped for node: [BatchNorm] bnhg1_low6_low1_branch2a
Warning: parameters not reshaped for node: [BatchNorm] bnhg1_low6_low1_branch2b
Warning: parameters not reshaped for node: [BatchNorm] bnhg1_low6_low1_branch2c
Warning: parameters not reshaped for node: [BatchNorm] bnhg1_low6_low2_branch2a
Warning: parameters not reshaped for node: [BatchNorm] bnhg1_low6_low2_branch2b
Warning: parameters not reshaped for node: [BatchNorm] bnhg1_low6_low2_branch2c
Warning: parameters not reshaped for node: [BatchNorm] bnhg1_low6_low5_branch2a
Warning: parameters not reshaped for node: [BatchNorm] bnhg1_low6_low5_branch2b
Warning: parameters not reshaped for node: [BatchNorm] bnhg1_low6_low5_branch2c
Warning: parameters not reshaped for node: [BatchNorm] bnhg1_low6_low6_low1_branch2a
Warning: parameters not reshaped for node: [BatchNorm] bnhg1_low6_low6_low1_branch2b
Warning: parameters not reshaped for node: [BatchNorm] bnhg1_low6_low6_low1_branch2c
Warning: parameters not reshaped for node: [BatchNorm] bnhg1_low6_low6_low2_branch2a
Warning: parameters not reshaped for node: [BatchNorm] bnhg1_low6_low6_low2_branch2b
Warning: parameters not reshaped for node: [BatchNorm] bnhg1_low6_low6_low2_branch2c
Warning: parameters not reshaped for node: [BatchNorm] bnhg1_low6_low6_low5_branch2a
Warning: parameters not reshaped for node: [BatchNorm] bnhg1_low6_low6_low5_branch2b
Warning: parameters not reshaped for node: [BatchNorm] bnhg1_low6_low6_low5_branch2c
Warning: parameters not reshaped for node: [BatchNorm] bnhg1_low6_low6_low6_low1_branch2a
Warning: parameters not reshaped for node: [BatchNorm] bnhg1_low6_low6_low6_low1_branch2b
Warning: parameters not reshaped for node: [BatchNorm] bnhg1_low6_low6_low6_low1_branch2c
Warning: parameters not reshaped for node: [BatchNorm] bnhg1_low6_low6_low6_low2_branch2a
Warning: parameters not reshaped for node: [BatchNorm] bnhg1_low6_low6_low6_low2_branch2b
Warning: parameters not reshaped for node: [BatchNorm] bnhg1_low6_low6_low6_low2_branch2c
Warning: parameters not reshaped for node: [BatchNorm] bnhg1_low6_low6_low6_low5_branch2a
Warning: parameters not reshaped for node: [BatchNorm] bnhg1_low6_low6_low6_low5_branch2b
Warning: parameters not reshaped for node: [BatchNorm] bnhg1_low6_low6_low6_low5_branch2c
Warning: parameters not reshaped for node: [BatchNorm] bnhg1_low6_low6_low6_low6_branch2a
Warning: parameters not reshaped for node: [BatchNorm] bnhg1_low6_low6_low6_low6_branch2b
Warning: parameters not reshaped for node: [BatchNorm] bnhg1_low6_low6_low6_low6_branch2c
Warning: parameters not reshaped for node: [BatchNorm] bnhg1_low6_low6_low6_low7_branch2a
Warning: parameters not reshaped for node: [BatchNorm] bnhg1_low6_low6_low6_low7_branch2b
Warning: parameters not reshaped for node: [BatchNorm] bnhg1_low6_low6_low6_low7_branch2c
Traceback (most recent call last):
  File "/home/mfujita/anaconda3/envs/mmdnn/bin/mmconvert", line 8, in <module>
    sys.exit(_main())
  File "/home/mfujita/anaconda3/envs/mmdnn/lib/python3.10/site-packages/mmdnn/conversion/_script/convert.py", line 102, in _main
    ret = convertToIR._convert(ir_args)
  File "/home/mfujita/anaconda3/envs/mmdnn/lib/python3.10/site-packages/mmdnn/conversion/_script/convertToIR.py", line 16, in _convert
    transformer = CaffeTransformer(args.network, args.weights, "tensorflow", inputshape[0], phase = args.caffePhase)
  File "/home/mfujita/anaconda3/envs/mmdnn/lib/python3.10/site-packages/mmdnn/conversion/caffe/transformer.py", line 336, in __init__
    graph = graph.transformed([DataReshaper({ # Reshape the parameters to TensorFlow's ordering
  File "/home/mfujita/anaconda3/envs/mmdnn/lib/python3.10/site-packages/mmdnn/conversion/caffe/graph.py", line 287, in transformed
    graph = transformer(graph)
  File "/home/mfujita/anaconda3/envs/mmdnn/lib/python3.10/site-packages/mmdnn/conversion/caffe/transformer.py", line 151, in __call__
    node.reshaped_data = weights.transpose(transpose_order)
ValueError: axes don't match array

opened by trombiano1 0

mmtoir from PyTorch 2 IR missing .json, .npy and .pb file

I'm trying to convert R101 from PyTorch to Tensorflow with script that mentioned below.

Input:
mmtoir -f pytorch -d Pytorch2IR --inputShape 3,112,112 -n R100PyTorch/Corrected.pth

Output:
PyTorch parser has not supported operator [onnx::Unsqueeze]. IR network strucuture may lost info.
IR network structure is saved as [Pytorch2IR.json].
IR network structure is saved as [Pytorch2IR.pb].
IR weights are saved as [Pytorch2IR.npy].

It's true that the weights and structures are saved but their size;

Pytorch2IR.json --> 1.5 kB Pytorch2IR.npy --> 7.3 kB Pytorch2IR.pb --> 188 bytes

The output of .json file as shown in below:

{
  "node": [
    {
      "attr": {
        "_output_shapes": {
          "list": {
            "shape": [
              {
                "dim": [
                  {
                    "size": "-1"
                  },
                  {
                    "size": "112"
                  },
                  {
                    "size": "112"
                  },
                  {
                    "size": "64"
                  }
                ]
              }
            ]
          }
        },
        "dilations": {
          "list": {
            "i": [
              "1",
              "1",
              "1",
              "1"
            ]
          }
        },
        "strides": {
          "list": {
            "i": [
              "1",
              "1",
              "1",
              "1"
            ]
          }
        },
        "use_bias": {
          "b": false
        },
        "pads": {
          "list": {
            "i": [
              "0",
              "1",
              "1",
              "0",
              "0",
              "1",
              "1",
              "0"
            ]
          }
        },
        "kernel_shape": {
          "list": {
            "i": [
              "3",
              "3",
              "3",
              "64"
            ]
          }
        },
        "group": {
          "i": "1"
        }
      },
      "op": "Conv",
      "name": "node926"
    }
  ]
}

Before trying convert from PyTorch2Tensorflow, I have done Mxnet2Tensorflow for R100 and R50. So, I am sure that mmtoir is working on my environment. I think there are some problem on mmtoir for Pytorch conversion. Anyone faced with that problem or solved?

opened by CengizhanYurdakul 0

Releases(0.3.1)

0.3.1(Jul 24, 2020)

Initial support for PyTorch1.5.1 .
Source code(tar.gz)
Source code(zip)
mmdnn-0.3.1-py2.py3-none-any.whl(311.40 KB)
0.3.0(May 3, 2020)
Updated dependencies：

NumPy 1.15.4 =>1.18.2

Pillow 5.2.0=>6.2.2

TensorFlow 1.13.1 =>1.15.2

Keras 2.2.4 =>2.2.5

MXNet 1.2.0 =>1.6.0

CNTK 2.6 =>2.7

SciPy==1.4.1

Scikit-image==0.16.2

Bug Fixed and Improvement:

Fix op [LRN]

Fix tensorflow_parser[Slice], [DepthwiseConv2dNative]

Fix caffe_emitter[Relu6]

Add op [Unpool]

Add keras [Affine]

Add tensorflow [AddV2], [FusedBatchNormV3]

Stability improvements.

Tips:

We recommand Python 3.6 since Python 2 has retired. If you experience incompatibility with other Python 3.x versions, please submit an issue.
Source code(tar.gz)
Source code(zip)
mmdnn-0.3.0-py2.py3-none-any.whl(312.69 KB)
0.2.5(Apr 15, 2019)
Updated dependencies:

Numpy 1.14.5 => 1.15.4 Tensorflow 1.9.0 => 1.13.1 Keras 2.1.6=>2.2.4 CoreML 0.8=> 2.1.0 Cntk 2.5.1=>2.6 MXNet 1.1.0 post0 => 1.2.0 Onnx 1.2.1=>1.4.1 Onnx-tf =>1.2.1(python3), 1.1.2(python2)

Bug Fixes and Improvements:

Fix Caffe constant emit

Add visual studio solution file

Add pytorch_emit [Prelu]

Add tf_parser & pytorch_emit [Cast]

Add keras_parser [Conv1DTranspose], [Conv2DTranspose], [Conv3DTranspose]

Source code(tar.gz)
Source code(zip)
0.2.4(Mar 3, 2019)
Initial implementation for RNN operators including GRU (Gated Recurrent Unit) cell, LSTM (Long Short-Term Memory) cell (no support for peephole yet) by matching the cell's corresponding pattern and fold them into a fuction or class in target framework.

Various fixes and stability improvements.

Source code(tar.gz)
Source code(zip)
mmdnn-0.2.4-py2.py3-none-any.whl(309.28 KB)
0.2.3(Mar 3, 2019)

Source code(tar.gz)
Source code(zip)
mmdnn-0.2.3-py2.py3-none-any.whl(277.35 KB)
0.2.2(Mar 3, 2019)
Support for for ssd and fastrcnn in tensorflow frozen parser.

Initial support for resnet-v2 , inception_resnet-v2 and facenet.

Add some ops which include Pack, Shape, Scale, StrideSlice, Prelu, Mul, Abs, Sub, Squeeze.

Source code(tar.gz)
Source code(zip)
mmdnn-0.2.2-py2.py3-none-any.whl(260.02 KB)
0.2.1(May 24, 2018)

Source code(tar.gz)
Source code(zip)
mmdnn-0.2.1-py2.py3-none-any.whl(244.49 KB)
0.2.0(May 16, 2018)

Formal release of CoreML parser and Caffe/Onnx/CoreML emitter. darknet parser initial release.
Source code(tar.gz)
Source code(zip)
mmdnn-0.2.0-py2.py3-none-any.whl(234.05 KB)
0.1.4(Mar 22, 2018)

Source code(tar.gz)
Source code(zip)
mmdnn-0.1.4-py2.py3-none-any.whl(200.51 KB)
0.1.3(Feb 26, 2018)

Source code(tar.gz)
Source code(zip)
mmdnn-0.1.3-py2.py3-none-any.whl(179.71 KB)
0.1.2(Dec 18, 2017)

Support Framework

Support Network

Models | Caffe | Keras | Tensorflow | CNTK | MXNet | PyTorch | :--------------------------------------------------:|:-----:|:-----:|:----------:|:----:|:-----:|:-------:| Inception V1 | √ | √ | √ | √ | √ Inception V3 | × | √ | √ | √ | √ ResNet V1 50 | × | √ | √ | o | √ | √ ResNet V2 152 | × | √ | √ | √ | √ | VGG 19 | √ | √ | √ | √ | √ | √ MobileNet_v1| × | √ | √ | × (No Relu6) | × Xception | × | √ | √ | × | × | SqueezeNet | | √ | √ | √ | √ |
Source code(tar.gz)
Source code(zip)
mmdnn-0.1.2-py2.py3-none-any.whl(127.89 KB)
0.1.1(Nov 29, 2017)

Support Framework

Support Network

Models | Caffe | Keras | Tensorflow | CNTK | MXNet | PyTorch | :--------------------------------------------------:|:-----:|:-----:|:----------:|:----:|:-----:|:-------:| Inception V1 | √ | √ | √ | √ | √ Inception V3 | × | √ | √ | √ | √ ResNet V1 50 | × | √ | √ | o | √ | √ ResNet V2 152 | × | √ | √ | √ | √ | VGG 19 | √ | √ | √ | √ | √ | √ MobileNet_v1| × | √ | √ | × (No Relu6) | × Xception | × | √ | √ | × | × | SqueezeNet | | √ | √ | √ | √ |
Source code(tar.gz)
Source code(zip)
mmdnn-0.1.1-py2.py3-none-any.whl(116.96 KB)

Owner

Microsoft

Open source projects and samples from Microsoft

GitHub

Robust Video Matting in PyTorch, TensorFlow, TensorFlow.js, ONNX, CoreML!

6.5k Jan 4, 2023

Robust Video Matting in PyTorch, TensorFlow, TensorFlow.js, ONNX, CoreML!

Robust Video Matting (RVM) English | 中文 Official repository for the paper Robust High-Resolution Video Matting with Temporal Guidance. RVM is specific

2 Aug 21, 2022

Very simple NCHW and NHWC conversion tool for ONNX. Change to the specified input order for each and every input OP. Also, change the channel order of RGB and BGR. Simple Channel Converter for ONNX.

scc4onnx Very simple NCHW and NHWC conversion tool for ONNX. Change to the specified input order for each and every input OP. Also, change the channel

16 Dec 22, 2022

tf2onnx - Convert TensorFlow, Keras and Tflite models to ONNX.

tf2onnx converts TensorFlow (tf-1.x or tf-2.x), tf.keras and tflite models to ONNX via command line or python api.

1.8k Jan 8, 2023

YOLOv5 in PyTorch > ONNX > CoreML > TFLite

This repository represents Ultralytics open-source research into future object detection methods, and incorporates lessons learned and best practices evolved over thousands of hours of training and evolution on anonymized client datasets. All code and models are under active development, and are subject to modification or deletion without notice.

34.1k Dec 31, 2022

YOLOv3 in PyTorch > ONNX > CoreML > TFLite

This repository represents Ultralytics open-source research into future object detection methods, and incorporates lessons learned and best practices

9.3k Jan 7, 2023

WHENet - ONNX, OpenVINO, TFLite, TensorRT, EdgeTPU, CoreML, TFJS, YOLOv4/YOLOv4-tiny-3L

HeadPoseEstimation-WHENet-yolov4-onnx-openvino ONNX, OpenVINO, TFLite, TensorRT, EdgeTPU, CoreML, TFJS, YOLOv4/YOLOv4-tiny-3L 1. Usage $ git clone htt

49 Sep 21, 2022

Automatic labeling, conversion of different data set formats, sample size statistics, model cascade

Simple Gadget Collection for Object Detection Tasks Automatic image annotation Conversion between different annotation formats Obtain statistical info

4 Aug 24, 2022

Microsoft Cognitive Toolkit (CNTK), an open source deep-learning toolkit

CNTK Chat Windows build status Linux build status The Microsoft Cognitive Toolkit (https://cntk.ai) is a unified deep learning toolkit that describes

17.3k Dec 29, 2022

Microsoft Cognitive Toolkit (CNTK), an open source deep-learning toolkit

CNTK Chat Windows build status Linux build status The Microsoft Cognitive Toolkit (https://cntk.ai) is a unified deep learning toolkit that describes

17k Feb 11, 2021

Convert Pytorch model to onnx or tflite, and the converted model can be visualized by Netron

5 Nov 19, 2022

A very simple tool to rewrite parameters such as attributes and constants for OPs in ONNX models. Simple Attribute and Constant Modifier for ONNX.

sam4onnx A very simple tool to rewrite parameters such as attributes and constants for OPs in ONNX models. Simple Attribute and Constant Modifier for

6 May 15, 2022

MMdnn is a set of tools to help users inter-operate among different deep learning frameworks. E.g. model conversion and visualization. Convert models between Caffe, Keras, MXNet, Tensorflow, CNTK, PyTorch Onnx and CoreML.

Related tags

Overview

MMdnn

Related Projects

Installation

Install manually

Install with docker image

Features

Support frameworks

Tested models

Usage

On-going frameworks

On-going Models

Examples

Official Tutorial

Users' Examples

Contributing

Intermediate Representation

Frameworks

Authors

Acknowledgements

License

Comments

Mxnet_model => IR format => Conversion code => Checkpoint => Tensorflow Model, Serving format

This is how the final structure of the converted model must look like in order to get served using TF-serving:

This is the accurate structure which TF-serving accept and generate predictions for.

Releases(0.3.1)

0.3.1(Jul 24, 2020)

0.3.0(May 3, 2020)

Updated dependencies：

Bug Fixed and Improvement:

Tips:

0.2.5(Apr 15, 2019)

Updated dependencies:

Bug Fixes and Improvements:

0.2.4(Mar 3, 2019)

0.2.3(Mar 3, 2019)

0.2.2(Mar 3, 2019)

0.2.1(May 24, 2018)

0.2.0(May 16, 2018)

0.1.4(Mar 22, 2018)

0.1.3(Feb 26, 2018)

0.1.2(Dec 18, 2017)

Support Framework

Support Network

0.1.1(Nov 29, 2017)

Support Framework

Support Network

Owner

Microsoft

Robust Video Matting in PyTorch, TensorFlow, TensorFlow.js, ONNX, CoreML!

Robust Video Matting in PyTorch, TensorFlow, TensorFlow.js, ONNX, CoreML!

Very simple NCHW and NHWC conversion tool for ONNX. Change to the specified input order for each and every input OP. Also, change the channel order of RGB and BGR. Simple Channel Converter for ONNX.

tf2onnx - Convert TensorFlow, Keras and Tflite models to ONNX.

YOLOv5 in PyTorch > ONNX > CoreML > TFLite

YOLOv3 in PyTorch > ONNX > CoreML > TFLite

WHENet - ONNX, OpenVINO, TFLite, TensorRT, EdgeTPU, CoreML, TFJS, YOLOv4/YOLOv4-tiny-3L

Automatic labeling, conversion of different data set formats, sample size statistics, model cascade

Microsoft Cognitive Toolkit (CNTK), an open source deep-learning toolkit

Microsoft Cognitive Toolkit (CNTK), an open source deep-learning toolkit

Convert Pytorch model to onnx or tflite, and the converted model can be visualized by Netron

A very simple tool to rewrite parameters such as attributes and constants for OPs in ONNX models. Simple Attribute and Constant Modifier for ONNX.

An executor that loads ONNX models and embeds documents using the ONNX runtime.

Simple tool to combine(merge) onnx models. Simple Network Combine Tool for ONNX.

ONNX-GLPDepth - Python scripts for performing monocular depth estimation using the GLPDepth model in ONNX

ONNX-PackNet-SfM: Python scripts for performing monocular depth estimation using the PackNet-SfM model in ONNX

Core ML tools contain supporting tools for Core ML model conversion, editing, and validation.

Matlab Python Heuristic Battery Opt - SMOP conversion and manual conversion

Convert onnx models to pytorch.