import copy import os.path from .other.module_online_data import GetOnlineData import time from .other.log import MyLogger import os from .remove_bg_pixian import RemoveBgPiXian from PIL import Image from .other.pic import Picture from .other.remove_bg_ali import RemoveBgALi import cv2 import numpy as np from middleware import UnicornException class Base(object): def __init__(self, image_data, lock, windows, num): self.lock = lock self.image_data = image_data self.num = num self.windows = windows self.get_online_data = GetOnlineData() self.file_path = image_data["file_path"] self.file_name = image_data["file_name"] self.file = os.path.split(self.file_path)[1] self.is_once_data = {} self.logger = MyLogger().logger def add_log(self, text, _type="info"): self.logger.info( "第{}个，图片名称：{},内容：{}".format(self.num, self.file, text) ) def show_image_info(self, data): data["file_path"] = self.file_path with self.lock: data = { "_type": "show_image_item_info", "data": data, } # self.windows.signal_data.emit(data) def send_info( self, text="", is_success=None, _type="show_text_browser", need_point_return=False, ): with self.lock: if is_success is not None: if is_success: processing_failed = 0 processing_successfully = 1 else: processing_failed = 1 processing_successfully = 0 # 分数返回 if need_point_return: # 分数返回 if self.is_once("add_point"): print( "第{}个，图片名称：{},内容：{}".format( self.num, self.file, "扣分返回" ) ) self.dispose_point(_type="add") self.refresh_times(cumulative_frequency_times_change=-1) pass # self.windows.signal_data.emit({"_type": "schedule", # "data": {"processing_failed": processing_failed, # "processing_successfully": processing_successfully, # }}) if text: data = { "_type": "show_text_browser", "data": text, } # self.windows.signal_data.emit(data) def check_path(self, _path): if not os.path.exists(_path): os.mkdir(_path) return True def is_once(self, key): if key not in self.is_once_data: self.is_once_data[key] = 0 return True return False def refresh_times(self, cumulative_frequency_times_change=0): data = { "_type": "refresh_times", "data": { "cumulative_frequency_times_change": cumulative_frequency_times_change }, } # self.windows.signal_data.emit(data) pass def dispose_point(self, _type): n = 3 while n: n -= 1 try: _r = self.get_online_data.dispose_point(_type) balance = _r["data"]["balance"] return True except: time.sleep(0.5) continue return False class DealOneImage(Base): def __init__(self, image_data, lock, windows, num): super().__init__(image_data, lock, windows, num) self.r_pixian = RemoveBgPiXian() def run(self, image_data, upload_pic_dict): self.file_path = image_data["file_path"] self.file = os.path.split(self.file_path)[1] self.root_path = image_data["root_path"] self.file_name = image_data["file_name"] # 直接调用抠图 # 1、增加获取key，2、key需要加密、3、429报错 重试再来拿一个KEY self.add_log("开始处理") remaining_times = self.get_online_data.get_cutout_image_times().get("balance") print("remaining_times", remaining_times) if remaining_times <= 0: raise UnicornException("次数不足，处理失败") # 检查图片上传是否有结束 n = 60 if self.file_path not in upload_pic_dict: # 不扣图 raise UnicornException("处理失败") s = time.time() image_deal_info = upload_pic_dict[self.file_path]["image_deal_info"] original_im = upload_pic_dict[self.file_path]["_im"] self.add_log("抠图中") # 抠图 out_root_path = "{}/已扣图".format(self.root_path) self.check_path(out_root_path) try: balance = self.get_online_data.get_cutout_image_times().get("balance") self.add_log("查询balance:{}成功".format(balance)) if balance <= 0: self.add_log("次数不足，处理失败") raise UnicornException("次数不足，处理失败") except: self.add_log("查询balance失败") raise UnicornException("查询balance失败") n = 0 second_cut_image = self.runPiXian(n, original_im=original_im) if second_cut_image is None: raise UnicornException("抠图失败") try: out_path = "{}/{}.png".format(out_root_path, self.file_name) if image_deal_info["二次抠图是否缩放"]: self.add_log(text="图片尺寸进行还原") original_im = image_deal_info["抠图扩边后PIL对象"] second_cut_image = self.picture_resize_to_original( second_cut_image, original_im ) # 创建空白图片并粘贴回去 _img_im = Image.new( mode="RGBA", size=image_deal_info["原始图片大小"], color=(0, 0, 0, 0) ) _img_im.paste( second_cut_image, box=( image_deal_info["抠图扩边后位置"][0], image_deal_info["抠图扩边后位置"][1], ), ) _img_im.save(out_path) except BaseException as e: # print(e) text = "{} 图片处理错误，代码44".format(e) self.add_log(text) self.send_info(text=text, is_success=False, need_point_return=True) _data = {"text": "出错/超时", "info": text, "status": False} return _data self.add_log(text="本张耗时：{}".format(time.time() - s)) self.send_info(text="抠图已完成", is_success=True) result = {"text": "已完成", "info": "", "status": True} return result def runPiXian(self, num, original_im): """执行pixian抠图""" num += 1 data = self.get_online_data.get_key_secret() key = (data["api_info"]["api_key"], data["api_info"]["api_serect"]) self.add_log("查询key成功") if not key: raise UnicornException("处理失败,请联系管理员") if self.is_once("sub_point"): # 调用扣分 self.refresh_times(cumulative_frequency_times_change=1) f = self.dispose_point(_type="sub") if not f: self.add_log(text="多次获取调用余额扣减失败") raise UnicornException("多次获取调用余额扣减失败") pixian_cutout_data = self.r_pixian.run_by_image_im(original_im, key) if pixian_cutout_data["status_code"] == 200: second_cut_image = pixian_cutout_data["im"] self.add_log(text="调用抠图完成") return second_cut_image elif pixian_cutout_data["status_code"] == 402: if num >= 2: self.cutoutFail(pixian_cutout_data) raise UnicornException("多次获取调用余额扣减失败") self.add_log( text="抠图失败:{}，延迟6秒".format(pixian_cutout_data["status_code"]) ) time.sleep(6) self.runPiXian(num, original_im=original_im) elif pixian_cutout_data["status_code"] == 429: if num >= 2: self.cutoutFail(pixian_cutout_data) return None self.add_log( text="抠图失败:{}，延迟10秒".format(pixian_cutout_data["status_code"]) ) time.sleep(10) self.runPiXian(num, original_im=original_im) else: _data = { "text": "出错/超时", "info": "抠图异常", } self.show_image_info(_data) self.get_online_data.dispose_point("add") if "message" in pixian_cutout_data: text = "抠图异常，code：{},message:{}".format( pixian_cutout_data["status_code"], pixian_cutout_data["message"] ) else: text = "抠图异常，code：{}".format(pixian_cutout_data["status_code"]) self.add_log(text) return None return second_cut_image def cutoutFail(self, pixian_cutout_data): """处理失败""" _data = { "text": "出错/超时", "info": "多次抠图失败:{}".format(pixian_cutout_data["status_code"]), } self.show_image_info(_data) self.add_log(text="多次抠图失败:{}".format(pixian_cutout_data["status_code"])) self.send_info( text="处理失败,请联系管理员", is_success=False, need_point_return=True, ) def picture_resize_to_original(self, _img, original_im): """ Parameters ---------- _img 需要还原的PIL对象 original_im 原图对象 Returns ------- """ # 将抠图结果转成mask # 将抠图结果放大到原始图大小 _img = _img.resize(original_im.size) new_big_mask = Image.new("RGB", _img.size, (0, 0, 0)) white = Image.new("RGB", _img.size, (255, 255, 255)) new_big_mask.paste(white, mask=_img.split()[3]) # ---------制作选区缩小的mask mask = cv2.cvtColor( np.asarray(new_big_mask), cv2.COLOR_BGR2GRAY ) # 将PIL 格式转换为 CV对象 mask[mask != 255] = 0 # 黑白反转 # mask = 255 - mask # 选区缩小10 kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (10, 10)) erode_im = cv2.morphologyEx(mask, cv2.MORPH_ERODE, kernel) # -------再进行抠图处理 mask = Image.fromarray( cv2.cvtColor(erode_im, cv2.COLOR_GRAY2RGBA) ) # CV 对象转 PIL transparent_im = Image.new("RGBA", original_im.size, (0, 0, 0, 0)) transparent_im.paste(original_im, (0, 0), mask.convert("L")) # 上述抠图结果进行拼接 _img.paste(transparent_im, (0, 0), transparent_im) return _img class DealOneImageBeforehand(Base): def __init__(self, image_data, lock, windows, num): super().__init__(image_data, lock, windows, num) self.r_ali = RemoveBgALi() def run(self, upload_pic_dict): # 增加阿里调用报错的重试机制 image_deal_info = {} try: cut_image, image_deal_info = self.get_image_cut() except BaseException as e: raise UnicornException("上传出错") upload_pic_dict[self.file_path] = { "image_deal_info": image_deal_info, "_im": cut_image, } return upload_pic_dict def get_image_cut(self): original_pic = Picture(self.file_path) original_pic.im = self.get_image_orientation(original_pic.im) original_pic.x, original_pic.y = original_pic.im.size original_pic.im = original_pic.im.convert("RGB") image_deal_info = {} image_deal_info["原始图片大小"] = (original_pic.x, original_pic.y) if original_pic.x * original_pic.y < 1000000: cut_image = original_pic.im image_deal_info["抠图扩边后图片大小"] = cut_image.size image_deal_info["二次抠图是否缩放"] = False image_deal_info["抠图扩边后位置"] = (0, 0, original_pic.x, original_pic.y) else: self.add_log("开始预抠图处理") cut_image = self.r_ali.get_image_cut( file_path=None, out_file_path=None, original_im=original_pic.im ) cut_image.save("XX1.png") self.add_log("预抠图处理结束") x1, y1, x2, y2 = cut_image.getbbox() image_deal_info["鞋子原始位置"] = (x1, y1, x2, y2) o_w, o_h = cut_image.size image_deal_info["鞋子原始抠图后大小"] = (o_w, o_h) # 扩边处理 _w, _h = x2 - x1, y2 - y1 out_px = 0.025 _w, _h = int(out_px * _w), int(out_px * _h) n_x1, n_y1, n_x2, n_y2 = x1 - _w, y1 - _h, x2 + _w, y2 + _h if n_x1 < 0: n_x1 = 0 if n_y1 < 0: n_y1 = 0 if n_x2 > o_w: n_x2 = o_w if n_y2 > o_h: n_y2 = o_h image_deal_info["抠图扩边后位置"] = (n_x1, n_y1, n_x2, n_y2) cut_image = original_pic.im.crop(image_deal_info["抠图扩边后位置"]) image_deal_info["抠图扩边后图片大小"] = cut_image.size x, y = image_deal_info["抠图扩边后图片大小"] # max_size = 32000000 max_size = 12000000 if x * y > max_size: r = max_size / (x * y) size = (int(x * r), int(y * r)) self.add_log( text="图片进行压缩，压缩前：{}，压缩后：{}".format( image_deal_info["抠图扩边后图片大小"], size ) ) image_deal_info["抠图扩边后PIL对象"] = copy.deepcopy(cut_image) cut_image = cut_image.resize(size=size) # print(cut_image.size) # print(image_deal_info["抠图扩边后PIL对象"].size) image_deal_info["二次抠图是否缩放"] = True else: image_deal_info["二次抠图是否缩放"] = False return cut_image, image_deal_info def get_image_cut_cloths(self): original_pic = Picture(self.file_path) original_pic.im = self.get_image_orientation(original_pic.im) original_pic.x, original_pic.y = original_pic.im.size original_pic.im = original_pic.im.convert("RGB") image_deal_info = {} image_deal_info["原始图片大小"] = (original_pic.x, original_pic.y) self.add_log("开始预抠图处理") remaining_times = self.get_online_data.get_cutout_image_times().get("balance") print("remaining_times", remaining_times) if remaining_times <= 0: raise UnicornException("次数不足，处理失败") self.get_online_data.dispose_point("sub") cut_images = self.r_ali.get_image_cut_cloths( file_path=None, out_file_path=None, original_im=original_pic.im ) if cut_images is None: self.get_online_data.dispose_point("add") self.add_log("预抠图处理结束") print("cut_images", cut_images) return x1, y1, x2, y2 = cut_image.getbbox() image_deal_info["鞋子原始位置"] = (x1, y1, x2, y2) o_w, o_h = cut_image.size image_deal_info["鞋子原始抠图后大小"] = (o_w, o_h) # 扩边处理 _w, _h = x2 - x1, y2 - y1 out_px = 0.025 _w, _h = int(out_px * _w), int(out_px * _h) n_x1, n_y1, n_x2, n_y2 = x1 - _w, y1 - _h, x2 + _w, y2 + _h if n_x1 < 0: n_x1 = 0 if n_y1 < 0: n_y1 = 0 if n_x2 > o_w: n_x2 = o_w if n_y2 > o_h: n_y2 = o_h image_deal_info["抠图扩边后位置"] = (n_x1, n_y1, n_x2, n_y2) cut_image = original_pic.im.crop(image_deal_info["抠图扩边后位置"]) image_deal_info["抠图扩边后图片大小"] = cut_image.size x, y = image_deal_info["抠图扩边后图片大小"] # max_size = 32000000 max_size = 12000000 if x * y > max_size: r = max_size / (x * y) size = (int(x * r), int(y * r)) self.add_log( text="图片进行压缩，压缩前：{}，压缩后：{}".format( image_deal_info["抠图扩边后图片大小"], size ) ) image_deal_info["抠图扩边后PIL对象"] = copy.deepcopy(cut_image) cut_image = cut_image.resize(size=size) # print(cut_image.size) # print(image_deal_info["抠图扩边后PIL对象"].size) image_deal_info["二次抠图是否缩放"] = True else: image_deal_info["二次抠图是否缩放"] = False return cut_image, image_deal_info def get_image_orientation(self, img): # 获取EXIF数据 exif = img._getexif() if exif is not None: # EXIF标签274对应的是Orientation orientation = exif.get(0x0112) if orientation == 2: # 水平翻转 img = img.transpose(Image.FLIP_LEFT_RIGHT) elif orientation == 3: # 旋转180度 img = img.rotate(180, expand=True) elif orientation == 4: # 垂直翻转 img = img.transpose(Image.FLIP_TOP_BOTTOM) elif orientation == 5: # 水平翻转后顺时针旋转90度 img = img.transpose(Image.FLIP_LEFT_RIGHT).transpose(Image.ROTATE_270) elif orientation == 6: # 顺时针旋转90度 img = img.transpose(Image.ROTATE_270) elif orientation == 7: # 水平翻转后逆时针旋转90度 img = img.transpose(Image.FLIP_LEFT_RIGHT).transpose(Image.ROTATE_90) elif orientation == 8: # 逆时针旋转90度 img = img.transpose(Image.ROTATE_90) else: print("没有EXIF数据或没有方向信息") orientation = 1 return img