使用PaddleClas（2.5）进行分类

说在前面：本人使用的是PPLCNetV2_base 模型进行分类的，其他的没有试过，不过感觉应该一样

首先

需要安装cuda和cudnn 这个就比较坑了，本人装了以后预测不准，可能装坏了，不过使用cpu也可以啊

先卸载之前的

cuda 的安装

CUDA Toolkit Archive | NVIDIA Developer

该网站提供你想要的版本，本人使用的是cuda11.7.0

选择自定义安装，去掉第二个

设置环境变量

Cudnn安装

cuDNN Archive | NVIDIA Developer

这个链接对应着许多版本，本人选择的是，许要注意cuda的版本号

将下载到的压缩包解压到 cuda 的安装路径C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.7下

安装paddle-gpu

pip install paddlepaddle-gpu==2.4.2

下载PaddleClas 源码

PaddlePaddle/PaddleClas: A treasure chest for visual classification and recognition powered by PaddlePaddle (github.com)

下载模型安装依赖：pip install --upgrade -r requirements.txt

运行

结果

将模型导出onnx

需要下载 Paddle2ONNX 模块，可以的话可以把源码下载下来有案例

# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import os
import sys
__dir__ = os.path.dirname(os.path.abspath(__file__))
sys.path.append(os.path.abspath(os.path.join(__dir__, '../')))

from ppcls.utils import config
from ppcls.engine.engine import Engine

import paddle
from paddle.static import InputSpec


if __name__ == "__main__":
    args = config.parse_args()

    args.config = r'D:\python\ai\PaddlePaddle\PaddleClas\PaddleClas-release-2.5\ppcls\configs\ImageNet\PPLCNetV2\PPLCNetV2_base.yaml'
    config = config.get_config(
        args.config, overrides=args.override, show=False)
    config.Global.pretrained_model = r'D:\python\ai\PaddlePaddle\PaddleClas\PaddleClas-release-2.5\weights\PPLCNetV2_base_pretrained'
    config.Infer.infer_imgs = r'D:\python\ai\PaddlePaddle\PaddleClas\PaddleClas-release-2.5\docs\images\inference_deployment\whl_demo.jpg'
    # config.Arch.pretrained = True
    config.Global.device = 'cpu'
    engine = Engine(config, mode="infer")
    layer=engine.model
    save_path = 'onnx.save/PPLCNetV2_base'
    x_spec = InputSpec([None, 3, 224, 224], 'float32', 'images')
    layer.eval()
    paddle.onnx.export(layer, save_path, input_spec=[x_spec])
    engine.infer()

结果

使用onnxruntime进行推理

import onnxruntime
import cv2
import torch
import numpy as np
import onnxruntime
import torch.nn.functional as F

x = np.random.random((2, 784)).astype('float32')

onnx_file = r'D:\python\ai\PaddlePaddle\PaddleClas\PaddleClas-release-2.5\tools\onnx.save\PPLCNetV2_base.onnx'

images = cv2.imread(r'D:\python\ai\PaddlePaddle\PaddleClas\PaddleClas-release-2.5\docs\images\inference_deployment\whl_demo.jpg', 1)

images=cv2.resize(images, (224, 224))
images = images[:, :, ::-1].transpose(2, 0, 1)

# images = np.ascontiguousarray(images)
images = images.astype('float32')
# images = torch.from_numpy(images).to("cpu")
# images = images.float()  # uint8 to fp16/32
images = images / 255.0  # 0 - 255 to 0.0 - 1.0

if len(images.shape) == 3:
    images = images[None]  # expand for batch dim


ort_sess = onnxruntime.InferenceSession(onnx_file, providers=['CPUExecutionProvider'])
ort_inputs = {ort_sess.get_inputs()[0].name: images}

ort_outs = ort_sess.run([ort_sess.get_outputs()[0].name], ort_inputs)
ort_outs=np.array(ort_outs, dtype=np.float32)
# ort_outs=ort_outs.float()
x=torch.from_numpy(ort_outs).to("cpu")
x = F.softmax(x,dim =-1)
# x = softmax(x)
x = x.numpy()
y = []
for idx, probs in enumerate(x):
    index =probs[0].argsort(axis=0)[-5:][::-1].astype(
        "int32") if not False else np.where(
            probs >= 0.5)[0].astype("int32")
    clas_id_list = []
    score_list = []
    label_name_list = []
    for i in index:
        clas_id_list.append(i.item())
        score_list.append(probs[0][i].item())

    result = {
        "class_ids": clas_id_list,
        "scores": np.around(
            score_list, decimals=5).tolist(),
    }
    # if file_names is not None:
    #     result["file_name"] = file_names[idx]
    if label_name_list is not None:
        result["label_names"] = label_name_list
    y.append(result)
    print(y)
# print(ort_outs)

结果：

备注：因为我转换图片比较粗暴，所以概率可能有点不一样

为解决问题：我的gpu运行不了，如何有可以直接使用的cuda和cudnn 包请给我来一份