钢铁表面缺陷检测
钢铁表面缺陷检测:数据：https://www.kaggle.com/c/severstal-steel-defect-detection/data
from pathlib import Path
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import cv2
class RLE:
"""
Encapsulates run-length-encoding functionality.
"""
MASK_H = 256
MASK_W = 1600
@classmethod
def from_str(cls, s):
if s != s:
return cls()
list_ =
return cls.from_list(list_)
@classmethod
def from_mask(cls, mask):
pixels = mask.T.flatten()
pixels = np.concatenate([, pixels, ])
runs = np.where(pixels != pixels[:-1]) + 1
runs -= runs[::2]
return cls.from_list(runs)
@classmethod
def from_list(cls, list_):
n_items = int(len(list_) / 2)
items = np.zeros((n_items, 2), dtype=np.uint64)
for i in range(n_items):
items = list_
items = list_
return cls(items)
def __init__(self, items=np.zeros((0, 0))):
self._items = items
@property
def items(self):
return self._items
def __iter__(self):
for idx, item in enumerate(self.items):
yield (item, item)# run, length
def __len__(self):
return self.items.shape
def to_mask(self):
mask = np.zeros(self.MASK_H * self.MASK_W, dtype=np.uint8)
for run, length in self:
run = int(run - 1)
end = int(run + length)
mask = 1
return mask.reshape(self.MASK_H, self.MASK_W, order='F')
def to_str_list(self):
list_ = []
for run, length in self:
list_.append(str(run))
list_.append(str(length))
return list_
def __str__(self):
if len(self) == 0:
return ''
return ' '.join(self.to_str_list())
def __repr__(self):
return self.__str__()
def make_mask(labels):
masks = np.zeros((256, 1600, 4), dtype=np.float32)
for c, label in enumerate(labels.values):
rle = RLE.from_str(label)
masks[:, :, c] = rle.to_mask()
return masks
img_size = 256 * 1600
train_csv = 'F:/sck/steel/train.csv'
df = pd.read_csv(train_csv)
df.head()
N_CLASSES = 4
rle_cols =
df['ImageId'], df['ClassId'] = zip(*df['ImageId_ClassId'].str.split('_'))
df['ClassId'] = df['ClassId'].astype(int)
df = df.pivot(index='ImageId', columns='ClassId', values='EncodedPixels')
df.columns = rle_cols
df['defects'] = df.count(axis=1)
df = df.loc
df.head()
r1 = df.loc['0007a71bf.jpg']
r1
m1 = make_mask(r1)[:, :, 2]
plt.imshow(m1)
for i in range(len(df)):
name=df.index
ri = df.loc
mi = make_mask(ri)
cv2.imwrite('F:/sck/steel/fourclass/1/'+name+'.png', mi[:,:,0])
cv2.imwrite('F:/sck/steel/fourclass/2/'+name+'.png', mi[:,:,1])
cv2.imwrite('F:/sck/steel/fourclass/3/'+name+'.png', mi[:,:,2])
cv2.imwrite('F:/sck/steel/fourclass/4/'+name+'.png', mi[:,:,3])
if(mi[:,:,0].max()>1):
print('11')
if(mi[:,:,1].max()>1):
print('11')
if(mi[:,:,2].max()>1):
print('11')
参考：【1】https://www.kaggle.com/lightforever/severstal-mlcomp-catalyst-infer-0-90672
【2】https://blog.csdn.net/HRLSW/article/details/102779214
【3】AI Studio优质数据集
【4】https://aistudio.baidu.com/aistudio/datasetdetail/10941
页:
[1]