[android][recipe] Making android native app using torchscript model with custom op #1041
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Fix allreduce implmentation in tutorial
Signed-off-by: Edward Z. Yang <ezyang@fb.com>
- Don't clean before running html-noplot (if you want to do a clean build, just run make clean yourself!) - Replace filename_pattern with ignore_pattern (with a negative lookahead assertion). This means that specifying a GALLERY_PATTERN will also skip parsing all Python files that you don't care about. This saves several seconds from the no-op doc build. - Use distutils to only copy rst files if the mtime is newer than the destination; this (probably) helps Sphinx avoid recompiling these files when it's useless. There are some semantic changes in this diff which is why I'm putting it up for review. Signed-off-by: Edward Z. Yang <ezyang@fb.com>
Reduce amount of unnecessary recompilation done by sphinx
jlin27
suggested changes
Jun 29, 2020
|
|
||
| - OpenCV SDK for Android | ||
|
|
||
| Our custom operator will be implemented using the OpenCV library. To use it for android, we need to download OpenCV SDK for Android with prebuilt libraries. |
| Making Android Application | ||
| ========================== | ||
|
|
||
| After we succeeded in having ``compute.pt``, we want to use this TorchScript model within android application. Using general TorchScript models (without custom operators) on Android, using java API, you can find `here <https://pytorch.org/mobile/android/>`_. We can not use this approach for our case, as our model uses a custom operator(``my_ops.warp_perspective``), default TorchScript execution will fail to find it. |
|
|
||
| After we succeeded in having ``compute.pt``, we want to use this TorchScript model within android application. Using general TorchScript models (without custom operators) on Android, using java API, you can find `here <https://pytorch.org/mobile/android/>`_. We can not use this approach for our case, as our model uses a custom operator(``my_ops.warp_perspective``), default TorchScript execution will fail to find it. | ||
|
|
||
| Registration of ops is not exposed to PyTorch java API, thus we need to build Android Application with native part (C++) and using LibTorch C++ API to implement and register the same custom operator for Android. As our operator uses the OpenCV library - we will use prebuilt OpenCV Android libraries and use the same functions from OpenCV. |
| Android Application Manifest | ||
| ============================ | ||
|
|
||
| Every android application has a manifest. |
| } | ||
| } | ||
|
|
||
| ``NativePeer#loadAndForwardModel`` is declared as ``native``, it does not have definition in java. Call to this method will beredispatched through jni to C++ method in our ``libpytorch_nativeapp.so``, in ``NativeApp/app/src/main/cpp/pytorch_nativeapp.cpp``. |
There was a problem hiding this comment.
Add space: "beredispatched" -> "be re-dispatched"
Capitalize "jni" -> "JNI"
|
|
||
|
|
||
|
|
||
| The full setup of this app you can find in `PyTorch Android Demo Application Repository <https://github.com/pytorch/android-demo-app/tree/master/NativeApp>`_. |
There was a problem hiding this comment.
Recommend to also add this line at the beginning of the recipe for users to download and follow along from the start.
jlin27
reviewed
Jun 30, 2020
| - How to use within this application TorchScript models with custom operators. | ||
|
|
||
| Setup | ||
| ===== |
There was a problem hiding this comment.
Nit: Recommend not to use the same header level === for sub-headers as is currently used for the title.
33e2a77 to
fc41a1d
Compare
fc41a1d to
8e4e379
Compare
|
Deploy preview for pytorch-tutorials-preview ready! Built with commit a79c227 https://deploy-preview-1041--pytorch-tutorials-preview.netlify.app |
jlin27
approved these changes
Jul 2, 2020
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
More condensed setup information and result application setup is here:
pytorch/android-demo-app#73