paddleDetection:OpenCV检测框转中文

注：OpenCV不能直接显示中文，通过PIL转换会损失一部分算力性能

修改源码（可视化）

./deploy/python/visualize.py

增加导入字体库和字体文件

from PIL import Image, ImageDraw, ImageFile, ImageFont

font_file = '/exp/work/video/PaddleDetection/deploy/pipeline/SourceHanSansCN-Medium.otf'

visualize_attr

def visualize_attr(im, results, boxes=None, is_mtmct=False):
    if isinstance(im, str):
        im = Image.open(im)
        im = np.ascontiguousarray(np.copy(im))
        im = cv2.cvtColor(im, cv2.COLOR_RGB2BGR)
    else:
        im = np.ascontiguousarray(np.copy(im))

    line_inter = im.shape[0] / 40.
    text_scale = max(0.5, im.shape[0] / 100.)
    # 将nparray图像转PIL图像
    im = Image.fromarray(im)

    for i, res in enumerate(results):
        print(i, res)
        if boxes is None:
            text_w = 3
            text_h = 1
        elif is_mtmct:
            box = boxes[i]  # multi camera, bbox shape is x,y, w,h
            text_w = int(box[0]) + 3
            text_h = int(box[1])
        else:
            box = boxes[i]  # single camera, bbox shape is 0, 0, x,y, w,h
            text_w = int(box[2]) + 3
            text_h = int(box[3])
        for text in res:
            text_h += int(line_inter)
            text_loc = (text_w, text_h)
            # 写入
            draw = ImageDraw.Draw(im)
            draw.text(
                text_loc,
                text,
                font=ImageFont.truetype(font_file, size=int(text_scale)), # 字体位置
                fill=(0, 255, 255))
    # 还原连续存储数组
    im = np.ascontiguousarray(np.copy(im))
    return im

visualize_vehicleplate

def visualize_vehicleplate(im, results, boxes=None):
    if isinstance(im, str):
        im = Image.open(im)
        im = np.ascontiguousarray(np.copy(im))
        im = cv2.cvtColor(im, cv2.COLOR_RGB2BGR)
    else:
        im = np.ascontiguousarray(np.copy(im))

    text_scale = max(1.0, im.shape[0] / 400.)
    line_inter = im.shape[0] / 40.
    # 将nparray图像转PIL图像
    im = Image.fromarray(im)
    for i, res in enumerate(results):
        if boxes is None:
            text_w = 3
            text_h = 1
        else:
            box = boxes[i]
            text = res
            if text == "":
                continue
            text_w = int(box[2])
            text_h = int(box[5] + box[3])
            text_loc = (text_w, text_h)
            # 写入
            draw = ImageDraw.Draw(im)
            draw.text(
                text_loc,
                text,
                font=ImageFont.truetype(font_file, size=int(text_scale)), # 字体位置
                fill=(0, 255, 255))
    # 还原连续存储数组
    im = np.ascontiguousarray(np.copy(im))
    return im

visualize_action

def visualize_action(im,
                     mot_boxes,
                     action_visual_collector=None,
                     action_text="",
                     video_action_score=None,
                     video_action_text=""):
    im = cv2.imread(im) if isinstance(im, str) else im
    im_h, im_w = im.shape[:2]

    text_scale = max(1, im.shape[1] / 40.)
    text_thickness = 2
    # 将nparray图像转PIL图像
    im = Image.fromarray(im)
    if action_visual_collector:
        id_action_dict = {}
        for collector, action_type in zip(action_visual_collector, action_text):
            id_detected = collector.get_visualize_ids()
            for pid in id_detected:
                id_action_dict[pid] = id_action_dict.get(pid, [])
                id_action_dict[pid].append(action_type)
        for mot_box in mot_boxes:
            # mot_box is a format with [mot_id, class, score, xmin, ymin, w, h] 
            if mot_box[0] in id_action_dict:
                text_position = (int(mot_box[3] + mot_box[5] * 0.75),
                                 int(mot_box[4] - 10))
                display_text = ', '.join(id_action_dict[mot_box[0]])

                draw = ImageDraw.Draw(im)
                draw.text(
                    text_position,
                    display_text,
                    font=ImageFont.truetype(size=int(text_scale)),  # 字体位置
                    fill=(0, 0, 255))

    if video_action_score:
        draw = ImageDraw.Draw(im)
        draw.text(
            (int(im_w / 2), int(15 * text_scale) + 5),
            video_action_text + ': %.2f' % video_action_score,
            font=ImageFont.truetype(font_file, size=int(text_scale)),  # 字体位置
            fill=(0, 0, 255))
    # 还原连续存储数组
    im = np.ascontiguousarray(np.copy(im))
    return im

修改源码（属性）

修改对应中文模块

./deploy/pipeline/

• pipeline.py

./deploy/pipeline/ppvehicle

• vehicle_attr.py
• vehicle_plate.py

./deploy/pipeline/pphuman

• attr_infer.py

执行脚本

# pphuman
python deploy/pipeline/pipeline.py --config deploy/pipeline/config/infer_cfg_pphuman.yml --video_file=demo_input/act1.mp4 --device=gpu --output_dir=demo_output

# ppvehicle
python deploy/pipeline/pipeline.py --config deploy/pipeline/config/infer_cfg_ppvehicle.yml --video_file=demo_input/car_t1.mp4 --device=gpu --output_dir=demo_output