Merge branch 'main' of github.com:microsoft/nn-Meter into main

README.md

@@ -1,7 +1,17 @@
-# Project
+# Introduction
 
-> This repo has been populated by an initial template to help get you started. Please
+>nn-Meter is a novel and efficient system to accurately predict the inference latency of DNN models on diverse edge devices. The key idea is dividing a whole model inference into kernels, i.e., the execution units of fused operators on a device, and conduct kernel-level prediction. 
-> make sure to update the content to build a great experience for community-building.
+nn-Meter contains two key techniques: (i) kernel detection to automatically detect the execution unit of model inference via a set of well-designed test cases; (ii) adaptive sampling to efficiently sample the most beneficial configurations from a large space to build accurate kernel-level latency predictors.
+nn-Meter currently evaluates four popular platforms on a large dataset of 26k models. It achieves 99.0% (mobile CPU), 99.1% (mobile Adreno 640 GPU), 99.0% (mobile Adreno 630 GPU), and 83.4% (Intel VPU) prediction accuracy.
+
+The current supported hardware and inference frameworks:
+
+| name |        Device       |    Framework   |    Processor   | +-10%  Accuracy |
+|:----:|:-------------------:|:--------------:|:--------------:|:------------------:|
+|  CPU |        Pixel4       |   TFLite v2.1  |  CortexA76 CPU |        99.0%       |
+|  GPU |         Mi9         |   TFLite v2.1  | Adreno 640 GPU |        99.1%       |
+| GPU1 |       Pixel3XL      |   TFLite v2.1  | Adreno 630 GPU |        99.0%       |
+|  VPU | Intel Movidius NCS2 | OpenVINO2019R2 |   Myriad VPU   |        83.4%       |
 
 As the maintainer of this project, please make a few updates:
 
@@ -10,6 +20,29 @@ As the maintainer of this project, please make a few updates:
 - Understanding the security reporting process in SECURITY.MD
 - Remove this section from the README
 
+## Installation
+
+To install nn-meter, please first install python3. The test environment uses anaconda python 3.6.10. Install the dependencies via: 
+`pip3 install -r requirements.txt`
+Please also check the versions of numpy, scikit_learn. The different versions may change the prediction accuracy of kernel predictors. 
+
+## Usage
+
+To run the latency predictor, we support two input formats. We include popular CNN models in `data/testmodels`
+
+#### 1. input model: xx.onnx or xx.pb:
+
+`python demo_with_converter.py --input_models data/testmodels/alexnet.onnx --mf alexnet`
+
+`python demo_with_converter.py --input_models data/testmodels/alexnet.pb --mf alexnet`
+
+It will firstly convert onnx and pb models into our defined IR json. We conduct kernel detection with the IR graph and predict kernel latency on the 4 measured edge devices. 
+
+#### 2. input model: the converted IR json:
+
+`python demo.py --input_models data/testmodels/alexnet_0.json --mf alexnet`
+
+
 ## Contributing
 
 This project welcomes contributions and suggestions.  Most contributions require you to agree to a