目錄

1. 整體流程

1）實體化模型

2）加載模型

3）輸入影像

4）輸出分類結果

5）完整代碼

2. 處理影像

1) How can i convert an RGB image into grayscale in Python?

2）PIL 處理影像的基本操作

3）影像通道數的理解

4）Convert 3 channel image to 2 channel

5）影像通道轉換

6）將所有的影像合并為一個numpy陣列

7）torch.from_numpy VS torch.Tensor

8）torch.squeeze() 和torch.unsqueeze()

3.issue

1）TypeError: 'module' object is not callable

2）TypeError: 'collections.OrderedDict' object is not callable

3) TypeError: __init__() missing 1 required positional argument: 'XX'

4) RuntimeError: Error(s) in loading state_dict for PythonNet: Missing key(s) in state_dict:

5) RuntimeError: Expected 4-dimensional input for 4-dimensional weight [128, 1, 3, 3], but got 2-dimensional input of size [480, 640] instead

6）RuntimeError: Input type (torch.FloatTensor) and weight type (torch.cuda.FloatTensor) should be the same or input should be a MKLDNN tensor and weight is a dense tensor

1. 整體流程

1）實體化模型

Assume that the content of YourClass.py is:

class YourClass:
    # ......

If you use:

from YourClassParentDir import YourClass  # means YourClass

from model import PythonNet

net= PythonNet(T=16).eval().cuda()

2）加載模型

import torch

net.load_state_dict(torch.load('checkpoint_max.pth'),False)

3）輸入影像

目的：從檔案夾中加載所有影像組合為一個 numpy array 作為模型輸入

原始影像輸入維度：（480,640）

目標影像輸入維度：（16,1,128,128）

import glob
from PIL import Image
import numpy as np
from torch.autograd import Variable

#獲取影像路徑
filelist = glob.glob('./testdata/a/*.jpg')

#打開影像open('frame_path')--》轉換為灰度圖convert('L')--》縮放影像resize((width, height)) --》合并檔案夾中的所有影像為一個numpy array
x = np.array([np.array(Image.open(frame).convert('L').resize((128,128))) for frame in filelist])

#用torch.from_numpy這個方法將numpy類轉換成tensor類
x = torch.from_numpy(x).type(torch.FloatTensor).cuda()

#擴充資料維度
x = Variable(torch.unsqueeze(x,dim=1).float(),requires_grad=False)

4）輸出分類結果

outputs = net(x)
_, predicted = torch.max(outputs,1)

torch.max()這個函式回傳輸入張量中所有元素的最大值，

回傳一個命名元組(values,indices)，其中values是給定維dim中輸入張量的每一行的最大值，indices是找到的每個最大值的索引位置(argmax)，也就是說，回傳的第一個值是對應影像在類別中的最大概率值，第二個值是最大概率值的對應類別，

Pytorch 分類問題輸出結果的資料整理方式：_, predicted = torch.max(outputs.data, 1) - stardsd - 博客園

5）完整代碼

from PIL import Image
from torch.autograd import Variable
import numpy as np
import torch
import glob
from model import PythonNet

##############處理輸入影像#######################################
#獲取影像路徑
filelist = glob.glob('./testdata/a/*.jpg')

#打開影像open('frame_path')--》轉換為灰度圖convert('L')--》縮放影像resize((width, height)) --》合并檔案夾中的所有影像為一個numpy array
x = np.array([np.array(Image.open(frame).convert('L').resize((128,128))) for frame in filelist])

#用torch.from_numpy這個方法將numpy類轉換成tensor類
x = torch.from_numpy(x).type(torch.FloatTensor).cuda()

#擴充資料維度
x = Variable(torch.unsqueeze(x,dim=1).float(),requires_grad=False)

#############定義預測函式#######################################
def predict(x):
    net= PythonNet(T=16).eval().cuda()
    net.load_state_dict(torch.load('checkpoint_max.pth'),False)

    outputs = net(x)
    _, predicted = torch.max(outputs,1)
    print("_:",_)
    print("predicted:",predicted)
    print("outputs:",outputs)

############輸入影像進行預測#####################################
predict(x)

2. 處理影像

1) How can i convert an RGB image into grayscale in Python?

matplotlib - How can I convert an RGB image into grayscale in Python? - Stack Overflow

2）PIL 處理影像的基本操作

python——PIL Image處理影像_aaon22357的博客-CSDN博客

3）影像通道數的理解

關于影像通道的理解 | TinaCristal's Blog

4）Convert 3 channel image to 2 channel

python - I have converted 3 channel RGB image into 2 channels grayscale image, How to decrease greyscale channels to 1? - Stack Overflow

5）影像通道轉換

影像通道轉換——從np.ndarray的[w, h, c]轉為Tensor的[c, w, h]_莫邪莫急的博客-CSDN博客

6）將所有的影像合并為一個numpy陣列

python — 如何在numpy陣列中加載多個影像？

7）torch.from_numpy VS torch.Tensor

torch.from_numpy VS torch.Tensor_麥克斯韋惡魔的博客-CSDN博客

8）torch.squeeze() 和torch.unsqueeze()

pytorch學習 中 torch.squeeze() 和torch.unsqueeze()的用法_xiexu911的博客-CSDN博客_torch.unsqueeze

3.issue

1）TypeError: 'module' object is not callable

python - TypeError: 'module' object is not callable - Stack Overflow

2）TypeError: 'collections.OrderedDict' object is not callable

pytorch加載模型報錯TypeError: ‘collections.OrderedDict‘ object is not callable_xiaoqiaoliushuiCC的博客-CSDN博客

3) TypeError: __init__() missing 1 required positional argument: 'XX'

Python成功解決TypeError: __init__() missing 1 required positional argument: ‘comment‘_肥鼠路易的博客-CSDN博客

4) RuntimeError: Error(s) in loading state_dict for PythonNet: Missing key(s) in state_dict:

pytorch加載模型報錯RuntimeError: Error(s) in loading state_dict for SSD:Missing key(s) in state_dict:... - 代碼先鋒網

5) RuntimeError: Expected 4-dimensional input for 4-dimensional weight [128, 1, 3, 3], but got 2-dimensional input of size [480, 640] instead

RuntimeError: Expected 4-dimensional input for 4-dimensional weight 64 3 3, but got 3-dimensional in_Steven_ycs的博客-CSDN博客

6）RuntimeError: Input type (torch.FloatTensor) and weight type (torch.cuda.FloatTensor) should be the same or input should be a MKLDNN tensor and weight is a dense tensor

RuntimeError: Input type (torch.cuda.FloatTensor) and weight type (torch.DoubleTensor) should be the_一千零一夜的博客-CSDN博客

??????