IntenseWebs/openvino
3
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Specification to ONNX external data feature (#2939)

Browse Source 
* Added external weights files doc

* review remarks

* review remarks

* review remarks

Co-authored-by: Michał Karzyński <4430709+postrational@users.noreply.github.com>

* review remarks. part 2

Co-authored-by: Michał Karzyński <4430709+postrational@users.noreply.github.com>

* fixed info about nullptr

* Apply suggestions from code review

Co-authored-by: Anastasiya Ageeva <anastasiya.ageeva@intel.com>

* review remarks. part 2

Co-authored-by: Michał Karzyński <4430709+postrational@users.noreply.github.com>
Co-authored-by: Anastasiya Ageeva <anastasiya.ageeva@intel.com>
This commit is contained in:
Mateusz Bencer 2020-11-17 04:57:57 +01:00 committed by GitHub
parent 0f539cc71c
commit 89f06586cf
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 36 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

33
docs/IE_DG/ONNX_Support.md
View File

@ -9,10 +9,41 @@ InferenceEngine::Core core;
auto network = core.ReadNetwork("model.onnx");
```
**Reshape feature:**
OpenVINO™ doesn't provide a mechanism to specify pre-processing (like mean values subtraction, reverse input channels) for the ONNX format.
If an ONNX model contains dynamic shapes for input, please use the `CNNNetwork::reshape` method for shape specialization.
Unsupported types of tensors:
**Weights saved in external files:**
OpenVINO™ supports ONNX models that store weights in external files. It is especially useful for models larger than 2GB because of protobuf limitations.
To read such models, use the `ReadNetwork` overload which takes `modelPath` as input parameter (both `std::string` and `std::wstring`).
Note that the `binPath` argument of `ReadNetwork` should be empty in this case, because paths to external weights are saved directly in an ONNX model.
Otherwise, a runtime exception is thrown.
Reading models with external weights is not supported by the `ReadNetwork(const std::string& model, const Blob::CPtr& weights)` overload.
Paths to external weight files are saved in an ONNX model; these paths are relative to the model's directory path.
It means that if a model is located at:
`home/user/workspace/models/model.onnx`
and a file that contains external weights:
`home/user/workspace/models/data/weights.bin`
the path saved in model should be:
`data/weights.bin`.
**NOTE**
* A single model can use many external weights files.
* Data of many tensors can be stored in a single external weights file (it is processed using offset and length values, which can be also saved in a model).
The described mechanism is the only possibility to read weights from external files. The following input parameters of the `ReadNetwork` function overloads are NOT supported for ONNX models and should be passed as empty:
* `const std::wstring& binPath`
* `const std::string& binPath`
* `const Blob::CPtr& weights`
You can find more details about external data mechanism in [ONNX documentation](https://github.com/onnx/onnx/blob/master/docs/ExternalData.md).
To convert a model to use external data feature, you can use [ONNX helpers functions](https://github.com/onnx/onnx/blob/master/onnx/external_data_helper.py).
**Unsupported types of tensors:**
* `string`,
* `complex64`,

4
docs/IE_DG/protecting_model_guide.md
View File

@ -43,6 +43,10 @@ Extensions](https://software.intel.com/en-us/sgx).
Use `InferenceEngine::Core::ReadNetwork()` to set model representations and
weights respectively.
Currently there are no possibility to read external weights from memory for ONNX models.
The `ReadNetwork(const std::string& model, const Blob::CPtr& weights)` function
should be called with `weights` passed as an empty `Blob`.
@snippet openvino/docs/snippets/protecting_model_guide.cpp part1
[deploy_encrypted_model]: img/deploy_encrypted_model.png