openvino/tools/pot/docs/E2eExample.md

End-to-end Command-line Interface Example

@sphinxdirective

This tutorial describes an example of running post-training quantization for MobileNet v2 model from PyTorch framework, particularly by the DefaultQuantization algorithm. The example covers the following steps:

• Environment setup
• Model preparation and converting it to the OpenVINO™ Intermediate Representation (IR) format
• Performance benchmarking of the original full-precision model
• Dataset preparation
• Accuracy validation of the full-precision model in the IR format
• Model quantization by the DefaultQuantization algorithm and accuracy validation of the quantized model
• Performance benchmarking of the quantized model

All the steps are based on the tools and samples of configuration files distributed with the Intel® Distribution of OpenVINO™ toolkit.

The example has been verified in Ubuntu 18.04 Operating System with Python 3.6 installed.

In case of issues while running the example, refer to :doc:POT Frequently Asked Questions <pot_docs_FrequentlyAskedQuestions> for help.

Model Preparation ####################

1. Navigate to <EXAMPLE_DIR>.

2. Download the MobileNet v2 PyTorch model using :doc:Model Downloader <omz_tools_downloader> tool from the Open Model Zoo repository:

   .. code-block:: sh

   omz_downloader --name mobilenet-v2-pytorch

   After that the original full-precision model is located in <EXAMPLE_DIR>/public/mobilenet-v2-pytorch/.

3. Convert the model to the OpenVINO™ Intermediate Representation (IR) format using :doc:Model Converter <omz_tools_downloader> tool:

   .. code-block:: sh

   omz_converter --name mobilenet-v2-pytorch

   After that the full-precision model in the IR format is located in <EXAMPLE_DIR>/public/mobilenet-v2-pytorch/FP32/.

For more information about the Model Optimizer, refer to its :doc:documentation <openvino_docs_MO_DG_Deep_Learning_Model_Optimizer_DevGuide>.

Performance Benchmarking of Full-Precision Models #################################################

Check the performance of the full-precision model in the IR format using :doc:Deep Learning Benchmark <openvino_inference_engine_tools_benchmark_tool_README> tool:

.. code-block:: sh

benchmark_app -m <EXAMPLE_DIR>/public/mobilenet-v2-pytorch/FP32/mobilenet-v2-pytorch.xml

Note that the results might be different dependently on characteristics of your machine. On a machine with Intel® Core™ i9-10920X CPU @ 3.50GHz it is like:

.. code-block:: sh

Latency: 4.14 ms Throughput: 1436.55 FPS

Dataset Preparation ####################

To perform the accuracy validation as well as quantization of a model, the dataset should be prepared. This example uses a real dataset called ImageNet.

To download images:

1. Go to the ImageNet <http://www.image-net.org/>__ homepage.
2. If you do not have an account, click the Signup button in the right upper corner, provide your data, and wait for a confirmation email.
3. Log in after receiving the confirmation email or if you already have an account. Go to the Download tab.
4. Select Download Original Images.
5. You will be redirected to the Terms of Access page. If you agree to the Terms, continue by clicking Agree and Sign.
6. Click one of the links in the Download as one tar file section.
7. Unpack the downloaded archive into <EXAMPLE_DIR>/ImageNet/.

Note that the registration process might be quite long.

Note that the ImageNet size is 50 000 images and takes around 6.5 GB of the disk space.

To download the annotation file:

1. Download archive <http://dl.caffe.berkeleyvision.org/caffe_ilsvrc12.tar.gz>__.
2. Unpack val.txt from the archive into <EXAMPLE_DIR>/ImageNet/.

After that the <EXAMPLE_DIR>/ImageNet/ dataset folder should have a lot of image files like ILSVRC2012_val_00000001.JPEG and the val.txt annotation file.

Accuracy Validation of Full-Precision Model in IR Format ########################################################

1. Create a new file in <EXAMPLE_DIR> and name it mobilenet_v2_pytorch.yaml. This is the Accuracy Checker configuration file.

2. Put the following text into mobilenet_v2_pytorch.yaml :

   .. code-block:: sh

   models: - name: mobilenet-v2-pytorch

      launchers:
        - framework: dlsdk
          device: CPU
          adapter: classification
   
      datasets:
        - name: classification_dataset
          data_source: ./ImageNet
          annotation_conversion:
            converter: imagenet
            annotation_file: ./ImageNet/val.txt
          reader: pillow_imread
   
          preprocessing:
            - type: resize
              size: 256
              aspect_ratio_scale: greater
              use_pillow: True
            - type: crop
              size: 224
              use_pillow: True
            - type: bgr_to_rgb
   
          metrics:
            - name: accuracy@top1
              type: accuracy
              top_k: 1
   
            - name: accuracy@top5
              type: accuracy
              top_k: 5
   

   where data_source: ./ImageNet is the dataset and annotation_file: ./ImageNet/val.txt is the annotation file prepared on the previous step. For more information about the Accuracy Checker configuration file refer to :doc:Accuracy Checker Tool documentation <omz_tools_accuracy_checker>.

3. Evaluate the accuracy of the full-precision model in the IR format by executing the following command in <EXAMPLE_DIR> :

   .. code-block:: sh

   accuracy_check -c mobilenet_v2_pytorch.yaml -m ./public/mobilenet-v2-pytorch/FP32/

   The actual result should be like 71.81% of the accuracy top-1 metric on VNNI based CPU. Note that the results might be different on CPUs with different instruction sets.

Model Quantization ####################

1. Create a new file in <EXAMPLE_DIR> and name it mobilenet_v2_pytorch_int8.json. This is the POT configuration file.

2. Put the following text into mobilenet_v2_pytorch_int8.json :

   .. code-block:: sh

   { "model": { "model_name": "mobilenet-v2-pytorch", "model": "./public/mobilenet-v2-pytorch/FP32/mobilenet-v2-pytorch.xml", "weights": "./public/mobilenet-v2-pytorch/FP32/mobilenet-v2-pytorch.bin" }, "engine": { "config": "./mobilenet_v2_pytorch.yaml" }, "compression": { "algorithms": [ { "name": "DefaultQuantization", "params": { "preset": "mixed", "stat_subset_size": 300 } } ] } }

   where "model": "./public/mobilenet-v2-pytorch/FP32/mobilenet-v2-pytorch.xml" and "weights": "./public/mobilenet-v2-pytorch/FP32/mobilenet-v2-pytorch.bin" specify the full-precision model in the IR format, "config": "./mobilenet_v2_pytorch.yaml" is the Accuracy Checker configuration file, and "name": "DefaultQuantization" is the algorithm name.

3. Perform model quantization by executing the following command in <EXAMPLE_DIR> :

   .. code-block:: sh

   pot -c mobilenet_v2_pytorch_int8.json -e

   The quantized model is placed into the subfolder with your current date and time in the name under the ./results/mobilenetv2_DefaultQuantization/ directory. The accuracy validation of the quantized model is performed right after the quantization. The actual result should be like 71.556% of the accuracy top-1 metric on VNNI based CPU. Note that the results might be different on CPUs with different instruction sets.

Performance Benchmarking of Quantized Model ###########################################

Check the performance of the quantized model using :doc:Deep Learning Benchmark <openvino_inference_engine_tools_benchmark_tool_README> tool:

.. code-block:: sh

benchmark_app -m <INT8_MODEL>

where <INT8_MODEL> is the path to the quantized model. Note that the results might be different dependently on characteristics of your machine. On a machine with Intel® Core™ i9-10920X CPU @ 3.50GHz it is like:

.. code-block:: sh

Latency: 1.54 ms Throughput: 3814.18 FPS

@endsphinxdirective