李沐的深度学习课 课程笔记

• 课程内容
• 课件
• github

S01E02 深度学习介绍

图片分类 image-net.org/

物体检测与分割 github.com/matterport/…

样式迁移 github.com/zhanghang19…

人脸合成

文件生成图片 openai.com/blog/dall-e

文字生成

大语言模型

无人驾驶

S01E03 环境搭建

创建环境

conda create -n d2l-zh python=3.8 pip

conda activate d2l-zh

安装依赖

pip install jupyter d2l torch torchvision

安装依赖过程中的报错

ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts. torchattacks 3.4.0 requires requests~=2.25.1, but you have requests 2.31.0 which is incompatible.

wget https://zh-v2.d2l.ai/d2l-zh.zip

unzip d2l-zh.zip

jupyter notebook

S01E04 pytorch 数据操作

向量访问

S01E05 pytorch 数据操作实现

import torch

x=torch.arange(12)
x

tensor([ 0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11])

x.shape

torch.Size([12])

X = x.reshape(3,4)
X

tensor([[ 0,  1,  2,  3],
        [ 4,  5,  6,  7],
        [ 8,  9, 10, 11]])

torch.zeros((2,3,4))

tensor([[[0., 0., 0., 0.],
         [0., 0., 0., 0.],
         [0., 0., 0., 0.]],

        [[0., 0., 0., 0.],
         [0., 0., 0., 0.],
         [0., 0., 0., 0.]]])

torch.tensor([[4,3,2,1],[2,3,4,5],[6,5,4,3]])

tensor([[4, 3, 2, 1],
        [2, 3, 4, 5],
        [6, 5, 4, 3]])

x=torch.tensor([1.0,2,3,4])
y=torch.tensor([2,2,2,2])
x+y,x-y,x*y,x/y,x**y

(tensor([3., 4., 5., 6.]),
 tensor([-1.,  0.,  1.,  2.]),
 tensor([2., 4., 6., 8.]),
 tensor([0.5000, 1.0000, 1.5000, 2.0000]),
 tensor([ 1.,  4.,  9., 16.]))

torch.exp(x)

tensor([ 2.7183,  7.3891, 20.0855, 54.5981])

x = torch.arange(12, dtype=torch.float32).reshape((3, 4))
y = torch.tensor([[2.0, 1, 4, 3], [1, 2, 3, 4], [4, 3, 2, 1]])
# 在0维进行连接 在2维矩阵中指行维度 所以此处是在行上连接（两个3行连接成6行）
torch.cat((x, y), dim=0)

tensor([[ 0.,  1.,  2.,  3.],
        [ 4.,  5.,  6.,  7.],
        [ 8.,  9., 10., 11.],
        [ 2.,  1.,  4.,  3.],
        [ 1.,  2.,  3.,  4.],
        [ 4.,  3.,  2.,  1.]])

# 两个4列连接成8列
torch.cat((x, y), dim=1)

tensor([[ 0.,  1.,  2.,  3.,  2.,  1.,  4.,  3.],
        [ 4.,  5.,  6.,  7.,  1.,  2.,  3.,  4.],
        [ 8.,  9., 10., 11.,  4.,  3.,  2.,  1.]])

x==y

tensor([[False,  True, False,  True],
        [False, False, False, False],
        [False, False, False, False]])

y.sum()

tensor(30.)

A = x.numpy()
B = torch.tensor(A)
type(A), type(B)

(numpy.ndarray, torch.Tensor)

a=torch.tensor([3.5])
a,a.item(),float(a)

(tensor([3.5000]), 3.5, 3.5)

使用 openvino 调用 onnx 模型

import cv2
import torchvision.transforms as transforms
from openvino.runtime import Core


def predict(onnx_file_path, image_path):
    img_bgr = cv2.imread(image_path, -1)
    PIC_RESIZE = (299, 299)
    img_bgr_resize = cv2.resize(img_bgr, PIC_RESIZE)
    img_rgb = cv2.cvtColor(img_bgr_resize, cv2.COLOR_BGR2RGB)

    data_transforms = transforms.Compose([
        transforms.ToTensor(),
        transforms.Resize(PIC_RESIZE, antialias=True),
        transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
    ])
    data = data_transforms(img_rgb)
    data.unsqueeze_(0)

    c_m = Core().compile_model(model=onnx_file_path, device_name='GPU')
    res_onnx = c_m([data])
    output_layer = c_m.output(0)
    res_onnx_i = res_onnx[output_layer]
    return res_onnx_i.argmax()


if __name__ == '__main__':
    pic_path = r"1716795604217.png"
    onnx_path = "sr_scene.onnx"
    class_list = ["setting", "map", "playing"]
    index = predict(onnx_path, pic_path)
    print(class_list[index])