~~tag>NPU VGG16 VIM4 Tensorflow Keras ~~

**Doc for version ddk-3.4.7.7**

====== VGG16 TensorFlow Keras VIM4 Demo 4 ======

{{indexmenu_n>4}}

[[https://www.google.com/search?q=VGG16|VGG16]] is a convolution neural net architecture that’s used for image recognition. It utilizes 16 layers with weights and is considered one of the best vision model architectures to date.

===== Get Source Code =====

[[gh>Daipuwei/Mini-VGG-CIFAR10]]


```shell
$ git clone https://github.com/Daipuwei/Mini-VGG-CIFAR10
```

===== Convert Model =====

==== Build virtual environment ====

Follow Docker official documentation to install Docker: [[https://docs.docker.com/engine/install/ubuntu/|Install Docker Engine on Ubuntu]].

Follow the script below to get Docker image:

```shell
docker pull numbqq/npu-vim4
```

==== Get convert tool ====

Download Tool from [[gl>khadas/vim4_npu_sdk]].

```shell
$ git lfs install
$ git lfs clone https://gitlab.com/khadas/vim4_npu_sdk.git
$ cd vim4_npu_sdk
$ ls
adla-toolkit-binary  adla-toolkit-binary-1.2.0.9  convert-in-docker.sh  Dockerfile  docs  README.md
```

  * ''adla-toolkit-binary/docs'' - SDK documentations
  * ''adla-toolkit-binary/bin'' - SDK tools required for model conversion
  * ''adla-toolkit-binary/demo'' - Conversion examples

<WRAP important>
If your kernel is older than 241129, please use version before tag ddk-3.4.7.7.
</WRAP>

==== Convert ====

We first need to convert the Keras model(''.h5'') into a TensorFlow model (''.pb''). We use this tool to convert [[gh>amir-abdi/keras_to_tensorflow]]

```shell
$ git clone https://github.com/amir-abdi/keras_to_tensorflow
```

Then we need to convert the TensorFlow model to an ADLA model (''.adla'')

Enter ''vim4_npu_sdk/demo'' and overwrite ''convert_adla.sh'' as follows.

```sh convert_adla.sh
#!/bin/bash
  
ACUITY_PATH=../bin/
#ACUITY_PATH=../python/tvm/
adla_convert=${ACUITY_PATH}adla_convert


if [ ! -e "$adla_convert" ]; then
    adla_convert=${ACUITY_PATH}adla_convert.py
fi

$adla_convert --model-type tensorflow \
        --model ./model_source/vgg16/vgg16.pb \
        --inputs image_input --input-shapes 32,32,3 \
        --outputs dense_2/Softmax \
        --inference-input-type float32 \
		--inference-output-type float32 \
        --quantize-dtype int8 --outdir tensorflow_output \
        --channel-mean-value "0,0,0,255" \
        --inference-input-type "float32" \
        --inference-output-type "float32" \
        --source-file vgg16_dataset.txt \
        --iterations 500 \
        --batch-size 1 \
        --target-platform PRODUCT_PID0XA003
```

Run ''convert_adla.sh'' to generate the VIM4 model. The converted model is ''xxx.adla'' in ''tensorflow_output''.

```shell
$ bash convert_adla.sh
```

===== Run NPU =====

==== Get source code ====

Clone the source code from our [[gh>khadas/vim4_npu_applications]].

```shell
$ git clone https://github.com/khadas/vim4_npu_applications
```

<WRAP important>
If your kernel is older than 241129, please use version before tag ddk-3.4.7.7.
</WRAP>

==== Install dependencies ====

```shell
$ sudo apt update
$ sudo apt install libopencv-dev python3-opencv cmake
```

==== Compile and run ====

=== Picture input demo ===

Put ''vgg16_int8.adla'' in ''vim4_npu_applications/vgg16/data/''.

```shell
# Compile
$ cd vim4_npu_applications/vgg16
$ mkdir build
$ cd build
$ cmake ..
$ make

# Run
$ ./vgg16 -m ../data/vgg16_int8.adla -p ../data/airplane.jpeg
```

{{:products:sbc:vim4:npu:demos:airplane.webp?400|}}

{{:products:sbc:vim4:npu:demos:vgg16-demo-output.webp?400|}}

<WRAP tip >
If your **VGG16** model classes are not the same as **CIFAR10**, please change ''data/vgg16_class.txt'' and the ''OBJ_CLASS_NUM'' in ''include/postprocess.h''.
</WRAP>