pySecureMR

Python bindings for SecureMR project.

Table of Contents

• Why pySecureMR
• Supported platform
• Install
  • Pip
  • Manual install
• Run test
• Supported operators

Why pySecureMR?

When developing a SecureMR app, it's not very easy to debug pipeline. You are not allowed to access each operator output directly. pySecureMR happens here to rescue you from complicated and painful debugging time. We bind most of SecureMR operators to python so you can call each operator and check input and output.

Supported platform

• Linux (ubuntu22): YES
• Windows (wsl2, ubuntu22): YES
• Mac: NO

Install

Pip

pip3 install securemr

Manual install

git clone https://github.com/Pico-Developer/pySecureMR
cd pySecureMR
pip3 install -e "."

Check installation:

python3 -c "import securemr"

Run test

pytest

Refer to test code to learn more about the usage.

Supported operators

ID	Name	Pybind
1	ARITHMETIC_COMPOSE	✅
4	ELEMENTWISE_MIN	✅
5	ELEMENTWISE_MAX	✅
6	ELEMENTWISE_MULTIPLY	✅
7	CUSTOMIZED_COMPARE	✅
8	ELEMENTWISE_OR	✅
9	ELEMENTWISE_AND	✅
10	ALL	✅
11	ANY	✅
12	NMS	✅
13	SOLVE_P_N_P	✅
14	GET_AFFINE	✅
15	APPLY_AFFINE	✅
16	APPLY_AFFINE_POINT	✅
17	UV_TO_3D_IN_CAM_SPACE	❌
18	ASSIGNMENT	✅
19	RUN_MODEL_INFERENCE	❌
21	NORMALIZE	✅
22	CAMERA_SPACE_TO_WORLD	❌
23	RECTIFIED_VST_ACCESS	❌
24	ARGMAX	✅
25	CONVERT_COLOR	✅
26	SORT_VEC	✅
27	INVERSION	✅
28	MAKE_TRANSFORM_MAT	✅
29	SORT_MAT	✅
30	SWITCH_GLTF_RENDER_STATUS	❌
31	UPDATE_GLTF	❌
32	RENDER_TEXT	❌
33	UPLOAD_TEXTURE_TO_GLTF	❌

Quick start for QNN

Install qnn. Only tested on Ubuntu22.04 & python3.10

python3 -m securemr.install_qnn

Run pytorch-to-qnn test and run on android device.

python3 -m securemr.test_pytorch_to_qnn

Visualize pipeline json

python3 -m securemr.viz.pipeline_viz path-to-pipeline.json

How to contribute

Before coding, please install develop related tools by:

make env

For new features, unittest is required.