zjsh_yolov11/推理图片反向上传CVAT/point/point_test.py

import cv2
import numpy as np
from ultralytics import YOLO
import os

# ====================== 用户配置 ======================
MODEL_PATH = 'best.pt'
IMAGE_SOURCE_DIR = '/media/hx/04e879fa-d697-4b02-ac7e-a4148876ebb0/dataset/1/20251230'
OUTPUT_DIR = './keypoints_txt'

IMG_EXTENSIONS = {'.png', '.jpg', '.jpeg', '.bmp', '.tiff', '.tif', '.webp'}

os.makedirs(OUTPUT_DIR, exist_ok=True)

# ====================== 保存函数 ======================
def save_yolo_kpts_with_bbox(save_path, cls_id, box_xyxy, kpts_xy, W, H):
    """
    追加写入一行 YOLO 格式：cls xc yc w h x1 y1 2 x2 y2 2 ...
    """
    x1, y1, x2, y2 = box_xyxy
    xc = (x1 + x2) / 2.0 / W
    yc = (y1 + y2) / 2.0 / H
    w = (x2 - x1) / W
    h = (y2 - y1) / H

    line_items = [str(cls_id), f"{xc:.6f}", f"{yc:.6f}", f"{w:.6f}", f"{h:.6f}"]

    for (x, y) in kpts_xy:
        x_norm = x / W
        y_norm = y / H
        line_items.append(f"{x_norm:.6f}")
        line_items.append(f"{y_norm:.6f}")
        line_items.append("2")  # visibility=2 表示可见

    with open(save_path, "a") as f:
        f.write(" ".join(line_items) + "\n")

    print(f"   💾 写入: {save_path}")


# ====================== 主程序 ======================
if __name__ == "__main__":
    print("🚀 开始 YOLO 检测框 + 关键点 TXT 输出")

    model = YOLO(MODEL_PATH)
    print(f"✅ 模型加载完成: {MODEL_PATH}")

    image_files = [
        f for f in os.listdir(IMAGE_SOURCE_DIR)
        if os.path.splitext(f.lower())[1] in IMG_EXTENSIONS
    ]

    if not image_files:
        print("❌ 未找到图像文件")
        exit(1)

    for img_filename in image_files:
        print(f"\n🖼️ 正在处理: {img_filename}")

        img_path = os.path.join(IMAGE_SOURCE_DIR, img_filename)
        img = cv2.imread(img_path)
        if img is None:
            print("❌ 无法读取图像")
            continue

        H, W = img.shape[:2]

        # 推理
        results = model(img)

        txt_name = os.path.splitext(img_filename)[0] + ".txt"
        save_path = os.path.join(OUTPUT_DIR, txt_name)

        # ✅ 关键修改：无论有没有检测结果，先创建（或清空）空文件
        open(save_path, 'w').close()  # 创建空文件，若存在则清空
        print(f"   📄 初始化空标签文件: {save_path}")

        has_detection = False
        for result in results:
            if result.boxes is None or len(result.boxes) == 0:
                continue

            boxes = result.boxes.xyxy.cpu().numpy()
            classes = result.boxes.cls.cpu().numpy()

            # 处理关键点（可能为 None）
            if hasattr(result, 'keypoints') and result.keypoints is not None:
                kpts_xy = result.keypoints.xy.cpu().numpy()  # (N, K, 2)
            else:
                kpts_xy = [np.array([])] * len(boxes)  # 无关键点时留空（但你的模型应有）

            for i in range(len(boxes)):
                cls_id = int(classes[i])
                box = boxes[i]
                kpts = kpts_xy[i] if len(kpts_xy) > i else np.array([])

                # 如果关键点数量不符合预期（比如不是4个），可选择跳过或填充
                # 这里假设模型输出固定数量关键点
                if kpts.size == 0:
                    print(f"   ⚠ 跳过无关键点的目标")
                    continue

                save_yolo_kpts_with_bbox(save_path, cls_id, box, kpts, W, H)
                has_detection = True

        # 可选：提示是否检测到内容
        if not has_detection:
            print(f"   ℹ️ 未检测到有效目标，保留空文件")

    print("\n==============================================")
    print("🎉 全部图像处理完毕！")
    print(f"📁 YOLO 输出目录：{OUTPUT_DIR}")
    print("✅ 所有图像均已生成对应 .txt 文件（含空文件）")
    print("==============================================")