CRNN_Chinese_Characters_Rec
CRNN_Chinese_Characters_Rec copied to clipboard
Published
20 hours ago •
Sierkinhane
Sierkinhane
训练不同长度标签
https://github.com/Sierkinhane/CRNN_Chinese_Characters_Rec/tree/stable/lib/dataset _360cc.py里面的getitem
如果有人成功尝试的话可以把代码贴出来 :）
w, h = image.size
if w / h < 280 / 32:
image_temp = image.resize((int(32 / h * w), 32), Image.BILINEAR)
w_temp, h_temp = image_temp.size
image_temp = cv2.cvtColor(np.asarray(image_temp), cv2.COLOR_RGB2BGR)
image_temp = cv2.copyMakeBorder(
image_temp, 0, 0, 0, 280 - w_temp, cv2.BORDER_CONSTANT, value=0)
res_image = Image.fromarray(cv2.cvtColor(image_temp, cv2.COLOR_BGR2RGB))
res_image = res_image.resize(self.size, Image.BILINEAR)
res_image = res_image.convert('L')
res_image = self.toTensor(res_image)
res_image.sub_(0.5).div_(0.5)
return res_image
if w / h > 280 / 32:
image_temp = image.resize((280, int(280 / w * h)), Image.BILINEAR)
w_temp, h_temp = image_temp.size
image_temp = cv2.cvtColor(np.asarray(image_temp), cv2.COLOR_RGB2BGR)
image_temp = cv2.copyMakeBorder(
image_temp, 0, 32 - h_temp, 0, 0, cv2.BORDER_CONSTANT, value=0)
res_image = Image.fromarray(cv2.cvtColor(image_temp, cv2.COLOR_BGR2RGB))
res_image = res_image.resize(self.size, Image.BILINEAR)
res_image = res_image.convert('L')
res_image = self.toTensor(res_image)
res_image.sub_(0.5).div_(0.5)
return res_image
if w / h == 280 / 32:
image = image.resize(self.size, Image.BILINEAR)
image = image.convert('L')
image = self.toTensor(image)
image.sub_(0.5).div_(0.5)
return image
如果有人成功尝试的话可以把代码贴出来 :）
w, h = image.size if w / h < 280 / 32: image_temp = image.resize((int(32 / h * w), 32), Image.BILINEAR) w_temp, h_temp = image_temp.size image_temp = cv2.cvtColor(np.asarray(image_temp), cv2.COLOR_RGB2BGR) image_temp = cv2.copyMakeBorder( image_temp, 0, 0, 0, 280 - w_temp, cv2.BORDER_CONSTANT, value=0) res_image = Image.fromarray(cv2.cvtColor(image_temp, cv2.COLOR_BGR2RGB)) res_image = res_image.resize(self.size, Image.BILINEAR) res_image = res_image.convert('L') res_image = self.toTensor(res_image) res_image.sub_(0.5).div_(0.5) return res_image if w / h > 280 / 32: image_temp = image.resize((280, int(280 / w * h)), Image.BILINEAR) w_temp, h_temp = image_temp.size image_temp = cv2.cvtColor(np.asarray(image_temp), cv2.COLOR_RGB2BGR) image_temp = cv2.copyMakeBorder( image_temp, 0, 32 - h_temp, 0, 0, cv2.BORDER_CONSTANT, value=0) res_image = Image.fromarray(cv2.cvtColor(image_temp, cv2.COLOR_BGR2RGB)) res_image = res_image.resize(self.size, Image.BILINEAR) res_image = res_image.convert('L') res_image = self.toTensor(res_image) res_image.sub_(0.5).div_(0.5) return res_image if w / h == 280 / 32: image = image.resize(self.size, Image.BILINEAR) image = image.convert('L') image = self.toTensor(image) image.sub_(0.5).div_(0.5) return image
您好，请问这个代码是把图片的读取方式换成PIL.Image了吗？ 还有这个是改_own.py里面的__getitem__函数吗？
如果有人成功尝试的话可以把代码贴出来 :）
w, h = image.size if w / h < 280 / 32: image_temp = image.resize((int(32 / h * w), 32), Image.BILINEAR) w_temp, h_temp = image_temp.size image_temp = cv2.cvtColor(np.asarray(image_temp), cv2.COLOR_RGB2BGR) image_temp = cv2.copyMakeBorder( image_temp, 0, 0, 0, 280 - w_temp, cv2.BORDER_CONSTANT, value=0) res_image = Image.fromarray(cv2.cvtColor(image_temp, cv2.COLOR_BGR2RGB)) res_image = res_image.resize(self.size, Image.BILINEAR) res_image = res_image.convert('L') res_image = self.toTensor(res_image) res_image.sub_(0.5).div_(0.5) return res_image if w / h > 280 / 32: image_temp = image.resize((280, int(280 / w * h)), Image.BILINEAR) w_temp, h_temp = image_temp.size image_temp = cv2.cvtColor(np.asarray(image_temp), cv2.COLOR_RGB2BGR) image_temp = cv2.copyMakeBorder( image_temp, 0, 32 - h_temp, 0, 0, cv2.BORDER_CONSTANT, value=0) res_image = Image.fromarray(cv2.cvtColor(image_temp, cv2.COLOR_BGR2RGB)) res_image = res_image.resize(self.size, Image.BILINEAR) res_image = res_image.convert('L') res_image = self.toTensor(res_image) res_image.sub_(0.5).div_(0.5) return res_image if w / h == 280 / 32: image = image.resize(self.size, Image.BILINEAR) image = image.convert('L') image = self.toTensor(image) image.sub_(0.5).div_(0.5) return image您好，请问这个代码是把图片的读取方式换成PIL.Image了吗？ 还有这个是改_own.py里面的__getitem__函数吗？
def getitem(self, idx):
@zcswdt 大大，不好意思，你的代码我在python3.6使用时，好像不太对，38 行h,w 应该不能直接承接 image.shape 吧，而且42行Image.size 返回的应该是hwc的一个数，怎么解包给w_temp 和 h_temp 呢？是我理解出错了吗？
@zcswdt 大大，不好意思，你的代码我在python3.6使用时，好像不太对，38 行h,w 应该不能直接承接 image.shape 吧，而且42行Image.size 返回的应该是h_w_c的一个数，怎么解包给w_temp 和 h_temp 呢？是我理解出错了吗？
这里读入的图片应该是灰度图。
@zcswdt good job
这块代码往哪个地方放
您好，你这个代码跟之前那个不一样，是有什么区别吗，用您这个方法怎么训练icdar2015数据集这种不定长的图片和标签吗，我放进去报错
