### Install Paddle Serving Docker Environment

Source: https://paddleclas.readthedocs.io/zh-cn/latest/extension/paddle_serving.html

Pull the Paddle Serving Docker image and run a container for GPU-accelerated serving. Access the container's bash shell for further setup.

```bash
nvidia-docker pull hub.baidubce.com/paddlepaddle/serving:0.2.0-gpu
nvidia-docker run -p 9292:9292 --name test -dit hub.baidubce.com/paddlepaddle/serving:0.2.0-gpu
nvidia-docker exec -it test bash
```

--------------------------------

### Start VisualDL for Monitoring

Source: https://paddleclas.readthedocs.io/zh-cn/latest/tutorials/getting_started.html

Launch VisualDL to monitor training progress in real-time, such as loss changes. Specify the log directory using --logdir.

```bash
visualdl --logdir ./scalar --host <host_IP> --port <port_num>

```

--------------------------------

### Install Paddle Serving CPU Server Package

Source: https://paddleclas.readthedocs.io/zh-cn/latest/extension/paddle_serving.html

Install the CPU version of the Paddle Serving server package if you intend to deploy a CPU-only service.

```python
pip install paddle-serving-server
```

--------------------------------

### Install Paddle Serving Python Packages

Source: https://paddleclas.readthedocs.io/zh-cn/latest/extension/paddle_serving.html

Install the necessary Paddle Serving Python client and server packages within the Docker container. Use a mirror for faster installation if needed.

```python
pip install paddlepaddle-gpu
pip install paddle-serving-client
pip install paddle-serving-server-gpu
```

--------------------------------

### Start Multi-GPU Training

Source: https://paddleclas.readthedocs.io/zh-cn/latest/tutorials/getting_started.html

Initiate multi-GPU and multi-process training using paddle.distributed.launch. Specify the GPUs to use with --selected_gpus and the configuration file with -c.

```python
# PaddleClas通过launch方式启动多卡多进程训练
# 通过设置FLAGS_selected_gpus 指定GPU运行卡号

python -m paddle.distributed.launch \
    --selected_gpus="0,1,2,3" \
    tools/train.py \
        -c ./configs/ResNet/ResNet50_vd.yaml

```

--------------------------------

### Training Script Configuration for Cutout

Source: https://paddleclas.readthedocs.io/zh-cn/latest/advanced_tutorials/image_augmentation/ImageAugment.html

Example of a training script configuration using the `run.sh` utility. This snippet shows how to specify a configuration file for the Cutout data augmentation strategy.

```bash
export PYTHONPATH=path_to_PaddleClas:$PYTHONPATH

python -m paddle.distributed.launch \
    --selected_gpus="0,1,2,3" \
    --log_dir=ResNet50_Cutout \
    tools/train.py \
        -c ./configs/DataAugment/ResNet50_Cutout.yaml
```

--------------------------------

### SSLD Distillation Configuration Example

Source: https://paddleclas.readthedocs.io/zh-cn/latest/advanced_tutorials/distillation/distillation.html

Configuration for distilling ResNeXt101_32x16d_wsl into ResNet50_vd using SSLD. It specifies the architecture name, pretrained model paths for teacher and student, and enables distillation.

```yaml
ARCHITECTURE:
    name: 'ResNeXt101_32x16d_wsl_distill_ResNet50_vd'
pretrained_model: "./pretrained/ResNeXt101_32x16d_wsl_pretrained/"
# pretrained_model:
#     - "./pretrained/ResNeXt101_32x16d_wsl_pretrained/"
#     - "./pretrained/ResNet50_vd_pretrained/"
use_distillation: True

```

--------------------------------

### Start Multi-GPU Training with Overrides

Source: https://paddleclas.readthedocs.io/zh-cn/latest/tutorials/getting_started.html

Start multi-GPU training and override configuration parameters using the -o flag. This allows for dynamic adjustments to training settings like mixed precision and VisualDL logging directory.

```python
python -m paddle.distributed.launch \
    --selected_gpus="0,1,2,3" \
    tools/train.py \
        -c ./configs/ResNet/ResNet50_vd.yaml \
        -o use_mix=1 \
	--vdl_dir=./scalar/

```

--------------------------------

### PaddleClas Distillation Training Command

Source: https://paddleclas.readthedocs.io/zh-cn/latest/advanced_tutorials/distillation/distillation.html

Command to launch the distillation training process. Replace `path_to_PaddleClas` with the actual path to your PaddleClas installation.

```bash
export PYTHONPATH=path_to_PaddleClas:$PYTHONPATH

python -m paddle.distributed.launch \
    --selected_gpus="0,1,2,3" \
    --log_dir=R50_vd_distill_MV3_large_x1_0 \
    tools/train.py \
        -c ./configs/Distillation/R50_vd_distill_MV3_large_x1_0.yaml

```

--------------------------------

### Start Paddle Serving HTTP Service

Source: https://paddleclas.readthedocs.io/zh-cn/latest/extension/paddle_serving.html

Launch the Paddle Serving HTTP prediction service using the exported model. Specify the model path, working directory, and the port for the service.

```python
python tools/serving/image_service_gpu.py serving/ppcls_model workdir 9292
```

--------------------------------

### Execute Prediction with Persistable Model

Source: https://paddleclas.readthedocs.io/zh-cn/latest/extension/paddle_inference.html

Run the inference engine to get predictions. This involves executing the compiled program with the input data and fetching the output.

```python
outputs = exe.run(infer_prog,
        feed={image.name: data},
        fetch_list=[out.name],
        return_numpy=False)
```

--------------------------------

### Download Paddle-Lite Benchmark Binary (ARMv7)

Source: https://paddleclas.readthedocs.io/zh-cn/latest/extension/paddle_mobile_inference.html

Download the pre-compiled benchmark binary for ARMv7 architecture. Use this if your Android device's ABI is v7.

```bash
wget -c https://paddle-inference-dist.bj.bcebos.com/PaddleLite/benchmark_0/benchmark_bin_v7
```

--------------------------------

### Executing the Training Script

Source: https://paddleclas.readthedocs.io/zh-cn/latest/advanced_tutorials/image_augmentation/ImageAugment.html

Command to execute the training script after setting up the environment and configuration.

```bash
sh tools/run.sh
```

--------------------------------

### Get Android CPU ABI

Source: https://paddleclas.readthedocs.io/zh-cn/latest/extension/paddle_mobile_inference.html

Execute this command on your Android device via ADB to determine its CPU architecture (ABI). This is crucial for selecting the correct pre-compiled benchmark binary.

```bash
adb shell getprop ro.product.cpu.abi
```

--------------------------------

### Configure and Fine-tune with SSLD Pre-trained Model

Source: https://paddleclas.readthedocs.io/zh-cn/latest/tutorials/quick_start.html

Sets up configuration for fine-tuning with an SSLD pre-trained model, recommending a smaller learning rate for intermediate layers. This approach further boosts accuracy.

```yaml
ARCHITECTURE:
    name: 'ResNet50_vd'
    params:
        lr_mult_list: [0.1, 0.1, 0.2, 0.2, 0.3]
pretrained_model: "./pretrained/ResNet50_vd_ssld_pretrained"
```

--------------------------------

### Execute Prediction with Paddle Predictor

Source: https://paddleclas.readthedocs.io/zh-cn/latest/extension/paddle_inference.html

Performs inference using a configured Paddle predictor. It involves getting input tensor names, preparing input data, and running the prediction.

```python
import numpy as np

input_names = predictor.get_input_names()
input_tensor = predictor.get_input_tensor(input_names[0])
input = np.random.randn(1, 3, 224, 224).astype("float32")
input_tensor.reshape([1, 3, 224, 224])
input_tensor.copy_from_cpu(input)
predictor.zero_copy_run()

```

--------------------------------

### Download Paddle-Lite Benchmark Binary (ARMv8)

Source: https://paddleclas.readthedocs.io/zh-cn/latest/extension/paddle_mobile_inference.html

Download the pre-compiled benchmark binary for ARMv8 architecture. This tool is used for evaluating model performance on Android devices.

```bash
wget -c https://paddle-inference-dist.bj.bcebos.com/PaddleLite/benchmark_0/benchmark_bin_v8
```

--------------------------------

### Download and Prepare Flowers102 Dataset

Source: https://paddleclas.readthedocs.io/zh-cn/latest/tutorials/quick_start.html

Download the flowers102 dataset and its labels, then extract the image archive.

```bash
cd dataset/flowers102
wget https://www.robots.ox.ac.uk/~vgg/data/flowers/102/102flowers.tgz
wget https://www.robots.ox.ac.uk/~vgg/data/flowers/102/imagelabels.mat
wget https://www.robots.ox.ac.uk/~vgg/data/flowers/102/setid.mat
tar -xf 102flowers.tgz
```

--------------------------------

### Execute Training Script

Source: https://paddleclas.readthedocs.io/zh-cn/latest/advanced_tutorials/distillation/distillation.html

Shell command to execute the training script after configuration.

```bash
sh tools/run.sh

```

--------------------------------

### Send Request to Paddle Serving HTTP Client

Source: https://paddleclas.readthedocs.io/zh-cn/latest/extension/paddle_serving.html

Send an image file to the running Paddle Serving HTTP service to get prediction results. Ensure the port number matches the service port.

```python
python tools/serving/image_http_client.py  9292 ./docs/images/logo.png
```

--------------------------------

### 安装Python依赖库

Source: https://paddleclas.readthedocs.io/zh-cn/latest/tutorials/install.html

安装PaddleClas所需的Python依赖库。这些依赖项在requirements.txt文件中列出。

```bash
pip install --upgrade -r requirements.txt

```

--------------------------------

### Convert Model using Command Line

Source: https://paddleclas.readthedocs.io/zh-cn/latest/extension/paddle_inference.html

A command-line utility to convert persistable models to inference models. Specify the model name, path to persistable weights, and output directory.

```bash
python tools/export_model.py \
    --m=模型名称 \
    --p=persistable 模型路径 \
    --o=model和params保存路径

```

--------------------------------

### Benchmark Optimized Model

Source: https://paddleclas.readthedocs.io/zh-cn/latest/extension/paddle_mobile_inference.html

Evaluate the performance of the optimized Paddle-Lite model using the benchmark binary. This command loads the optimized model from the specified directory.

```bash
bash benchmark.sh ./benchmark_bin_v8 ./opt_models result_armv8.txt
```

--------------------------------

### Train with SSLD Pre-trained Model

Source: https://paddleclas.readthedocs.io/zh-cn/latest/tutorials/quick_start.html

Executes the training process using the SSLD pre-trained model configuration, aiming for higher accuracy on the flowers102 dataset.

```bash
export CUDA_VISIBLE_DEVICES=0
python -m paddle.distributed.launch \
      --selected_gpus="0" \
      tools/train.py \
        -c ./configs/quick_start/ResNet50_vd_ssld_finetune.yaml
```

--------------------------------

### Generate Dataset Label Files

Source: https://paddleclas.readthedocs.io/zh-cn/latest/tutorials/quick_start.html

Create label files for training, validation, and testing sets using the provided script. Combines train and extra lists for enhanced distillation.

```bash
python generate_flowers102_list.py jpg train > train_list.txt
python generate_flowers102_list.py jpg valid > val_list.txt
python generate_flowers102_list.py jpg test > extra_list.txt
cat train_list.txt extra_list.txt > train_extra_list.txt
```

--------------------------------

### Train from Scratch with ResNet50_vd

Source: https://paddleclas.readthedocs.io/zh-cn/latest/tutorials/quick_start.html

Initiates training of a ResNet50_vd model without using any pre-trained weights. This is useful for understanding baseline performance.

```bash
export CUDA_VISIBLE_DEVICES=0
python -m paddle.distributed.launch \
      --selected_gpus="0" \
      tools/train.py \
        -c ./configs/quick_start/ResNet50_vd.yaml
```

--------------------------------

### Model Optimization using opt_mac Tool

Source: https://paddleclas.readthedocs.io/zh-cn/latest/extension/paddle_mobile_inference.html

Optimize a Paddle inference model using the Paddle-Lite opt tool on macOS. This process generates an optimized model file (e.g., .nb) suitable for mobile deployment.

```bash
model_file="../MobileNetV1/model"
param_file="../MobileNetV1/params"
opt_models_dir="./opt_models"
mkdir ${opt_models_dir}
./opt_mac --model_file=${model_file} \
    --param_file=${param_file} \
    --valid_targets=arm \
    --optimize_out_type=naive_buffer \
    --prefer_int8_kernel=false \
    --optimize_out=${opt_models_dir}/MobileNetV1
```

--------------------------------

### Run Model Speed Benchmark (ARMv8)

Source: https://paddleclas.readthedocs.io/zh-cn/latest/extension/paddle_mobile_inference.html

Execute the benchmark script to evaluate model inference speed on the connected Android device. This command also includes an option to optimize the model before benchmarking.

```bash
sh tools/lite/benchmark.sh ./benchmark_bin_v8 ./inference result_armv8.txt true
```

--------------------------------

### 模型推理评估脚本

Source: https://paddleclas.readthedocs.io/zh-cn/latest/models/models_intro.html

用于模型推理和性能基准测试的Bash脚本。需要设置PYTHONPATH并指定模型文件、参数文件及模型名称。

```bash
#!/usr/bin/env bash

export PYTHONPATH=$PWD:$PYTHONPATH

python tools/infer/predict.py \
    --model_file='pretrained/infer/model' \
    --params_file='pretrained/infer/params' \
    --enable_benchmark=True \
    --model_name=ResNet50_vd \
    --use_tensorrt=True \
    --use_fp16=False \
    --batch_size=1

```

--------------------------------

### Navigate to PaddleClas Directory

Source: https://paddleclas.readthedocs.io/zh-cn/latest/tutorials/quick_start.html

Change the current directory to the root of the PaddleClas project.

```bash
cd path_to_PaddleClas
```

--------------------------------

### Return to PaddleClas Root Directory

Source: https://paddleclas.readthedocs.io/zh-cn/latest/tutorials/quick_start.html

Navigate back to the main PaddleClas directory after dataset preparation.

```bash
cd ../../
```

--------------------------------

### Perform Knowledge Distillation Training

Source: https://paddleclas.readthedocs.io/zh-cn/latest/tutorials/quick_start.html

Executes the knowledge distillation training process using the configured settings. This method leverages a teacher model and unlabeled data to improve the student model's accuracy.

```bash
export CUDA_VISIBLE_DEVICES=0
python -m paddle.distributed.launch \
      --selected_gpus="0" \
      tools/train.py \
        -c ./configs/quick_start/R50_vd_distill_MV3_large_x1_0.yaml
```

--------------------------------

### 安装GPU版本PaddlePaddle

Source: https://paddleclas.readthedocs.io/zh-cn/latest/tutorials/install.html

使用pip安装最新GPU版本的PaddlePaddle。请确保已安装CUDA、cuDNN和NCCL运行环境。

```bash
pip install paddlepaddle-gpu --upgrade

```

--------------------------------

### 验证PaddlePaddle安装

Source: https://paddleclas.readthedocs.io/zh-cn/latest/tutorials/install.html

运行此Python代码以检查PaddlePaddle是否成功安装。建议在安装PaddlePaddle后执行此操作。

```python
import paddle.fluid as fluid
fluid.install_check.run_check()

```

--------------------------------

### 安装特定版本VisualDL

Source: https://paddleclas.readthedocs.io/zh-cn/latest/tutorials/install.html

如果VisualDL安装失败，请尝试使用此命令安装指定版本的VisualDL。

```bash
pip3 install --upgrade visualdl==2.0.0b3 -i https://mirror.baidu.com/pypi/simple

```

--------------------------------

### Download Pre-trained Models

Source: https://paddleclas.readthedocs.io/zh-cn/latest/tutorials/quick_start.html

Download specified pre-trained models for ResNet50_vd, ResNet50_vd_ssld, and MobileNetV3_large_x1_0. Models are downloaded to the specified path and decompressed.

```bash
python tools/download.py -a ResNet50_vd -p ./pretrained -d True
python tools/download.py -a ResNet50_vd_ssld -p ./pretrained -d True
python tools/download.py -a MobileNetV3_large_x1_0 -p ./pretrained -d True
```

--------------------------------

### Configure Paddle Predictor with AnalysisConfig

Source: https://paddleclas.readthedocs.io/zh-cn/latest/extension/paddle_inference.html

Sets up the prediction configuration using AnalysisConfig for optimized inference. Enables GPU, disables logging, enables IR optimization, and configures TensorRT.

```python
from paddle.fluid.core import AnalysisConfig
from paddle.fluid.core import create_paddle_predictor
config = AnalysisConfig(model文件路径, params文件路径)
config.enable_use_gpu(8000, 0)
config.disable_glog_info()
config.switch_ir_optim(True)
config.enable_tensorrt_engine(
        precision_mode=AnalysisConfig.Precision.Float32,
        max_batch_size=1)

# no zero copy方式需要去除fetch feed op
config.switch_use_feed_fetch_ops(False)

predictor = create_paddle_predictor(config)

```

--------------------------------

### Build Inference Engine with Inference Model

Source: https://paddleclas.readthedocs.io/zh-cn/latest/extension/paddle_inference.html

Load the inference model structure and weights directly to build the inference engine. This method is used when the model already contains structural information.

```python
import fluid

place = fluid.CPUPlace()
exe = fluid.Executor(place)
[program, feed_names, fetch_lists] = fluid.io.load_inference_model(
        模型的存储路径,
        exe,
        model_filename=保存的模型文件,
        params_filename=保存的参数文件)
compiled_program = fluid.compiler.CompiledProgram(program)
```

--------------------------------

### Build Inference Engine with Persistable Model

Source: https://paddleclas.readthedocs.io/zh-cn/latest/extension/paddle_inference.html

Construct the network structure and load weights to build the inference engine when using a persistable model. Ensure necessary imports and configurations are in place.

```python
import fluid
from ppcls.modeling.architectures.resnet_vd import ResNet50_vd

place = fluid.CPUPlace()
exe = fluid.Executor(place)
startup_prog = fluid.Program()
infer_prog = fluid.Program()
with fluid.program_guard(infer_prog, startup_prog):
    with fluid.unique_name.guard():
        image = create_input()
        image = fluid.data(name='image', shape=[None, 3, 224, 224], dtype='float32')
        out = ResNet50_vd.net(input=input, class_dim=1000)
infer_prog = infer_prog.clone(for_test=True)
fluid.load(program=infer_prog, model_path=persistable 模型路径, executor=exe)
```

--------------------------------

### 克隆PaddleClas模型库

Source: https://paddleclas.readthedocs.io/zh-cn/latest/tutorials/install.html

克隆PaddleClas模型库到本地。请先导航到您希望存放代码的目录。

```bash
cd path_to_clone_PaddleClas
git clone https://github.com/PaddlePaddle/PaddleClas.git

```

--------------------------------

### Dataset List Format

Source: https://paddleclas.readthedocs.io/zh-cn/latest/tutorials/data.html

Format for 'train_list.txt' and 'val_list.txt' files, separating image path and label with a space.

```text
# 每一行采用"空格"分隔图像路径与标注

ILSVRC2012_val_00000001.JPEG 65
...

```

--------------------------------

### Export Paddle Serving Model

Source: https://paddleclas.readthedocs.io/zh-cn/latest/extension/paddle_serving.html

Use the provided script to export a trained model into the format required by Paddle Serving. Specify the model name, pretrained path, and output directory.

```python
python tools/export_serving_model.py -m ResNet50_vd -p ./pretrained/ResNet50_vd_pretrained/ -o serving
```

--------------------------------

### Predict with Inference Model

Source: https://paddleclas.readthedocs.io/zh-cn/latest/extension/paddle_inference.html

Execute predictions using the inference model format. This command requires specifying the image path, model directory, model file, parameters file, and GPU usage.

```bash
python tools/infer/py_infer.py \
    --i=图片路径 \
    --d=模型的存储路径 \
    --m=保存的模型文件 \
    --p=保存的参数文件 \
    --use_gpu=True
```

--------------------------------

### Train MobileNetV3 Large Model

Source: https://paddleclas.readthedocs.io/zh-cn/latest/tutorials/quick_start.html

Trains a MobileNetV3_large_x1_0 model, demonstrating the use of different architectures. This allows comparison of performance based on computational and storage needs.

```bash
export CUDA_VISIBLE_DEVICES=0
python -m paddle.distributed.launch \
      --selected_gpus="0" \
      tools/train.py \
        -c ./configs/quick_start/MobileNetV3_large_x1_0_finetune.yaml
```

--------------------------------

### Using AutoAugment for Image Augmentation

Source: https://paddleclas.readthedocs.io/zh-cn/latest/advanced_tutorials/image_augmentation/ImageAugment.html

Demonstrates how to apply the AutoAugment image augmentation policy in PaddleClas. This method involves decoding, resizing, and then applying a randomly selected sub-policy from a predefined set of transformations.

```python
from ppcls.data.imaug import DecodeImage
from ppcls.data.imaug import ResizeImage
from ppcls.data.imaug import ImageNetPolicy
from ppcls.data.imaug import transform

size = 224

decode_op = DecodeImage()
resize_op = ResizeImage(size=(size, size))
autoaugment_op = ImageNetPolicy()

ops = [decode_op, resize_op, autoaugment_op]

imgs_dir = 图像路径
fnames = os.listdir(imgs_dir)
for f in fnames:
    data = open(os.path.join(imgs_dir, f)).read()
    img = transform(data, ops)
```

--------------------------------

### Run Inference using Command Line

Source: https://paddleclas.readthedocs.io/zh-cn/latest/extension/paddle_inference.html

Executes model inference using the predict.py script. Specify input image, model, and parameters files, along with GPU and TensorRT options.

```bash
python ./tools/infer/predict.py \
    -i=./test.jpeg \
    -m=./resnet50-vd/model \
    -p=./resnet50-vd/params \
    --use_gpu=1 \
    --use_tensorrt=True

```

--------------------------------

### Flowers102 Data Structure

Source: https://paddleclas.readthedocs.io/zh-cn/latest/tutorials/data.html

Organize Flowers102 data with an 'jpg' directory containing images and 'train_list.txt' and 'val_list.txt' files.

```text
PaddleClas/dataset/flowers102/
|_ jpg/
|  |_ image_03601.jpg
|  |_ ...
|  |_ image_02355.jpg
|_ train_list.txt
|_ val_list.txt
```

--------------------------------

### 查看PaddlePaddle版本

Source: https://paddleclas.readthedocs.io/zh-cn/latest/tutorials/install.html

使用此命令快速查看已安装的PaddlePaddle版本。适用于快速检查环境配置。

```python
python -c "import paddle; print(paddle.__version__)"

```