PaddleClas图像分类项目

PaddleClas

PaddleClas

训练环境搭建

https://www.paddlepaddle.org.cn/documentation/docs/zh/install/docker/linux-docker.html

# 拉取docker镜像
sudo docker pull registry.baidubce.com/paddlepaddle/paddle:2.4.2-gpu-cuda11.2-cudnn8.2-trt8.0

# 官方镜像
sudo docker run -id --name face --gpus all --shm-size=64G -p 33331:22 -p 9100:9100 -p 9101:9101  registry.baidubce.com/paddlepaddle/paddle:2.4.2-gpu-cuda11.2-cudnn8.2-trt8.0 bash

# 自己做的镜像
sudo docker run -id --name folds --gpus all --shm-size=32G -p 33332:22 -p 9102:9102 -p 9103:9103 face_shape_cls:1.0 bash

sudo docker exec -it folds bash

# 开启ssh，远程调试用
service ssh start
service ssh status

可忽略的报错：

ERROR: onnx 1.13.1 has requirement protobuf<4,>=3.20.2, but you'll have protobuf 3.20.0 which is incompatible.

训练模型和推理

修改配置文件

# global configs
Global:
  checkpoints: null
  pretrained_model: null
  output_dir: ./output/
  device: gpu
  save_interval: 100
  eval_during_train: True
  eval_interval: 1
  epochs: 300
  print_batch_step: 10
  use_visualdl: False
  # used for static mode and model export
  image_shape: [3, 224, 224]
  save_inference_dir: ./inference

# model architecture
Arch:
  name: ResNet50_vd
  class_num: 3

# loss function config for traing/eval process
Loss:
  Train:
    - CELoss:
        weight: 1.0
  Eval:
    - CELoss:
        weight: 1.0


Optimizer:
  name: Momentum
  momentum: 0.9
  lr:
    name: Cosine
    learning_rate: 0.0125
#    learning_rate: 0.05
    warmup_epoch: 5
  regularizer:
    name: 'L2'
    coeff: 0.00002


# data loader for train and eval
DataLoader:
  Train:
    dataset:
      name: ImageNetDataset
      image_root: ./dataset/folds/
      cls_label_path: ./dataset/folds/train_list.txt
      transform_ops:
        - DecodeImage:
            to_rgb: True
            channel_first: False
        - RandCropImage:
            size: 224
        - RandFlipImage:
            flip_code: 1
        - NormalizeImage:
            scale: 1.0/255.0
            mean: [0.485, 0.456, 0.406]
            std: [0.229, 0.224, 0.225]
            order: ''

    sampler:
      name: DistributedBatchSampler
      batch_size: 16
      drop_last: False
      shuffle: True
    loader:
      num_workers: 2
      use_shared_memory: True

  Eval:
    dataset: 
      name: ImageNetDataset
      image_root: ./dataset/folds/
      cls_label_path: ./dataset/folds/val_list.txt
      transform_ops:
        - DecodeImage:
            to_rgb: True
            channel_first: False
        - ResizeImage:
            resize_short: 256
        - CropImage:
            size: 224
        - NormalizeImage:
            scale: 1.0/255.0
            mean: [0.485, 0.456, 0.406]
            std: [0.229, 0.224, 0.225]
            order: ''
    sampler:
      name: DistributedBatchSampler
      batch_size: 32
      drop_last: False
      shuffle: False
    loader:
      num_workers: 2
      use_shared_memory: True

Infer:
  infer_imgs: docs/images/inference_deployment/whl_demo.jpg
  batch_size: 10
  transforms:
    - DecodeImage:
        to_rgb: True
        channel_first: False
    - ResizeImage:
        resize_short: 256
    - CropImage:
        size: 224
    - NormalizeImage:
        scale: 1.0/255.0
        mean: [0.485, 0.456, 0.406]
        std: [0.229, 0.224, 0.225]
        order: ''
    - ToCHWImage:
  PostProcess:
    name: Topk
    topk: 1
    class_id_map_file: ./dataset/folds/label_list.txt

Metric:
  Train:
    - TopkAcc:
        topk: [1, 1]
  Eval:
    - TopkAcc:
        topk: [1, 1]

训练命令

python tools/train.py -c ./ppcls/configs/quick_start/ResNet50_vd.yaml -o Arch.pretrained=True

训练后会在 PaddleClas/output/ 目录下生成模型文件

部分结果：

CELoss: 0.08261, loss: 0.08261, top1: 0.98000


[2023/06/20 13:36:26] ppcls INFO: [Eval][Epoch 100][Iter: 0/4]CELoss: 0.07598, loss: 0.07598, top1: 0.96875, batch_cost: 0.66089s, reader_cost: 0.56775, ips: 48.41936 images/sec
[2023/06/20 13:36:27] ppcls INFO: [Eval][Epoch 100][Avg]CELoss: 0.03693, loss: 0.03693, top1: 0.99000
[2023/06/20 13:36:27] ppcls INFO: Already save model in ./output/ResNet50_vd/best_model
[2023/06/20 13:36:27] ppcls INFO: [Eval][Epoch 100][best metric: 0.9899999499320984]

推理命令

# 黑眼圈推理命令
python tools/infer.py -c black_ResNet50.yaml -o Infer.infer_imgs=dataset/black/black/2000855.jpg -o Global.pretrained_model=output/ResNet50_vd/best_model
# 法令纹推理命令
python tools/infer.py -c folds_ResNet50.yaml -o Infer.infer_imgs=dataset/folds/folds/2000358.png -o Global.pretrained_model=output/ResNet50_vd/best_model

服务化部署

分类模型导出

参考：https://github.com/PaddlePaddle/PaddleClas/blob/release/2.5/docs/zh_CN/deployment/export_model.md

https://blog.csdn.net/loutengyuan/article/details/126674945

# 黑眼圈导出推理模型
python tools/export_model.py \
    -c black_ResNet50.yaml \
    -o Global.pretrained_model=output/ResNet50_vd/best_model \
    -o Global.save_inference_dir=./deploy/model/black_ResNet50_infer

# 法令纹导出推理模型
python tools/export_model.py \
    -c folds_ResNet50.yaml \
    -o Global.pretrained_model=output/ResNet50_vd/best_model \
    -o Global.save_inference_dir=./deploy/model/folds_ResNet50_infer

就能在你的保存路径中得到inference 模型。

用 paddle_serving_client 命令把下载的 inference 模型转换成易于 Server 部署的模型格式：

cd deploy/model
# 转换 ResNet50_vd 模型
# 黑眼圈
python3.7 -m paddle_serving_client.convert \
--dirname ./black_ResNet50_infer/ \
--model_filename inference.pdmodel  \
--params_filename inference.pdiparams \
--serving_server ./black_ResNet50_vd_serving/ \
--serving_client ./black_ResNet50_vd_client/

# 法令纹
python3.7 -m paddle_serving_client.convert \
--dirname ./folds_ResNet50_infer/ \
--model_filename inference.pdmodel  \
--params_filename inference.pdiparams \
--serving_server ./folds_ResNet50_vd_serving/ \
--serving_client ./folds_ResNet50_vd_client/

启动服务命令

python3.7 folds_classification_web_service.py &>log.txt &
python3.7 black_classification_web_service.py &>log.txt &

# 停止服务
python3.7 -m paddle_serving_server.serve stop