convert_dataset_colmap.py

import os
import shutil
import argparse
import logging
import struct
import numpy as np
from collections import namedtuple

# 設定日誌
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')

# COLMAP 檔案讀寫的輔助結構
Camera = namedtuple("Camera", ["id", "model", "width", "height", "params"])
Image = namedtuple("Image", ["id", "qvec", "tvec", "camera_id", "name", "xys", "point3D_ids"])
Point3D = namedtuple("Point3D", ["id", "xyz", "rgb", "error", "image_ids", "point2D_idxs"])

def read_next_bytes(fid, num_bytes, format_char_sequence, endian_character="<"):
    return struct.unpack(endian_character + format_char_sequence, fid.read(num_bytes))

def read_images_binary(path):
    images = {}
    with open(path, "rb") as fid:
        num_reg_images = read_next_bytes(fid, 8, "Q")[0]
        for _ in range(num_reg_images):
            binary_image_properties = read_next_bytes(fid, 64, "i4d3di")
            image_id = binary_image_properties[0]
            qvec = np.array(binary_image_properties[1:5])
            tvec = np.array(binary_image_properties[5:8])
            camera_id = binary_image_properties[8]
            
            image_name = ""
            current_char = read_next_bytes(fid, 1, "c")[0]
            while current_char != b"\x00":
                image_name += current_char.decode("utf-8")
                current_char = read_next_bytes(fid, 1, "c")[0]
            
            num_points2D = read_next_bytes(fid, 8, "Q")[0]
            fid.seek(24 * num_points2D, 1)
            
            images[image_id] = Image(
                id=image_id, qvec=qvec, tvec=tvec,
                camera_id=camera_id, name=image_name,
                xys=np.empty((0, 2)), point3D_ids=np.empty(0))
    return images

def write_images_binary(path, images):
    with open(path, "wb") as fid:
        fid.write(struct.pack("<Q", len(images)))
        for _, image in images.items():
            fid.write(struct.pack("<i", image.id))
            fid.write(struct.pack("<4d", *image.qvec))
            fid.write(struct.pack("<3d", *image.tvec))
            fid.write(struct.pack("<i", image.camera_id))
            fid.write(image.name.encode("utf-8") + b"\x00")
            fid.write(struct.pack("<Q", len(image.xys)))

def run_command(cmd):
    logging.info(f"執行命令: {cmd}")
    exit_code = os.system(cmd)
    if exit_code != 0:
        logging.error(f"命令執行失敗，返回碼: {exit_code}。腳本終止。")
        exit(exit_code)
    logging.info("命令執行成功。")

def run_full_colmap_and_split(data_path, colmap_arg=""):
    logging.info(f"===== 開始對 {data_path} 執行全自動 COLMAP 重建與分割 =====")
    colmap_command = f'"{colmap_arg}"' if len(colmap_arg) > 0 else "colmap"

    # --- 步驟 1: 準備統一的臨時工作目錄 ---
    logging.info("步驟 1: 準備統一的工作目錄...")
    work_dir = os.path.join(data_path, "colmap_work_dir")
    image_dir = os.path.join(work_dir, "images")
    
    shutil.rmtree(work_dir, ignore_errors=True)
    os.makedirs(image_dir)

    train_image_dir = os.path.join(data_path, "train", "images")
    test_image_dir = os.path.join(data_path, "test", "images")

    for d in [os.path.join(data_path, "train_data"), os.path.join(data_path, "test_data")]:
        if os.path.exists(os.path.join(d, "input")):
            os.rename(os.path.join(d, "input"), os.path.join(d, "images"))

    train_files = set(os.listdir(train_image_dir))
    test_files = set(os.listdir(test_image_dir))

    for f in train_files: shutil.copy(os.path.join(train_image_dir, f), image_dir)
    for f in test_files: shutil.copy(os.path.join(test_image_dir, f), image_dir)
    logging.info(f"已將 {len(train_files)} 個訓練影像和 {len(test_files)} 個測試影像複製到工作目錄。")

    # --- 步驟 2: 執行完整的 COLMAP 流程 ---
    logging.info("步驟 2: 執行完整的 COLMAP 流程...")
    db_path = os.path.join(work_dir, "database.db")
    
    # BUG FIX: 在 feature_extractor 命令中強制指定 PINHOLE 相機模型
    cmd_feature = (f'{colmap_command} feature_extractor '
                   f'--database_path "{db_path}" '
                   f'--image_path "{image_dir}" '
                   f'--ImageReader.single_camera 1 '
                   f'--ImageReader.camera_model PINHOLE '
                   f'--SiftExtraction.max_num_features 8192 '
                   f'--SiftExtraction.upright 0') # <-- 核心修改！
    run_command(cmd_feature)
    
    cmd_matcher = f'{colmap_command} exhaustive_matcher --database_path "{db_path}"'
    run_command(cmd_matcher)
    
    sparse_dir = os.path.join(work_dir, "sparse")
    os.makedirs(sparse_dir)
    cmd_mapper = (f'{colmap_command} mapper '
                  f'--database_path "{db_path}" '
                  f'--image_path "{image_dir}" '
                  f'--output_path "{sparse_dir}" '
                  f'--Mapper.ba_refine_focal_length 0 '
                  f'--Mapper.ba_refine_principal_point 0 '
                  f'--Mapper.ba_refine_extra_params 0 '
                  f'--Mapper.min_num_matches 4 '
                  f'--Mapper.init_min_num_inliers 4 '
                  f'--Mapper.abs_pose_max_error 12.0 '
                  f'--Mapper.abs_pose_min_num_inliers 4 '
                  f'--Mapper.init_max_forward_motion 0.95 '
                  f'--Mapper.init_min_tri_angle 4.0 '
                  f'--Mapper.multiple_models 0')
    
    # 尝试标准重建
    exit_code = os.system(cmd_mapper)
    if exit_code != 0:
        logging.warning("标准重建失败，尝试更宽松的参数...")
        # 更宽松的重建参数
        cmd_mapper_fallback = (f'{colmap_command} mapper '
                              f'--database_path "{db_path}" '
                              f'--image_path "{image_dir}" '
                              f'--output_path "{sparse_dir}" '
                              f'--Mapper.ba_refine_focal_length 0 '
                              f'--Mapper.ba_refine_principal_point 0 '
                              f'--Mapper.ba_refine_extra_params 0 '
                              f'--Mapper.min_num_matches 2 '
                              f'--Mapper.init_min_num_inliers 2 '
                              f'--Mapper.abs_pose_max_error 20.0 '
                              f'--Mapper.abs_pose_min_num_inliers 2 '
                              f'--Mapper.init_max_forward_motion 0.99 '
                              f'--Mapper.init_min_tri_angle 2.0 '
                              f'--Mapper.multiple_models 0')
        run_command(cmd_mapper_fallback)

    # --- 步驟 3: 分割 COLMAP 模型 ---
    logging.info("步驟 3: 分割 COLMAP 模型...")
    unified_model_path = os.path.join(sparse_dir, "0")
    if not os.path.exists(unified_model_path):
        subdirs = [d for d in os.listdir(sparse_dir) if os.path.isdir(os.path.join(sparse_dir, d)) and d.isdigit()]
        if len(subdirs) == 1:
            unified_model_path = os.path.join(sparse_dir, subdirs[0])
            logging.info(f"找到了 COLMAP 輸出模型於: {unified_model_path}")
        else:
            logging.error("COLMAP mapper 未能成功生成唯一的 sparse 模型文件夾，腳本終止。")
            return

    images_data = read_images_binary(os.path.join(unified_model_path, "images.bin"))

    train_images = {}
    test_images = {}
    for img_id, img in images_data.items():
        if img.name in train_files:
            train_images[img_id] = img
        elif img.name in test_files:
            test_images[img_id] = img

    logging.info(f"分割完成: {len(train_images)} 個訓練影像，{len(test_images)} 個測試影像。")
    if len(train_images) == 0 or len(test_images) == 0:
        logging.warning("警告：訓練集或測試集中的影像未能全部成功註冊，分割後可能為空。")

    # --- 步驟 4: 創建最終的輸出目錄結構 ---
    logging.info("步驟 4: 創建最終的輸出目錄...")
    for split, split_images in [("train", train_images), ("test", test_images)]:
        if not split_images: # 如果分割後沒有圖片，則跳過
            logging.warning(f"{split} 中沒有成功註冊的影像，跳過生成 sparse 文件。")
            continue
        output_dir = os.path.join(data_path, split)
        output_sparse_dir = os.path.join(output_dir, "sparse", "0")  # 直接創建 sparse/0 目錄
        shutil.rmtree(output_sparse_dir, ignore_errors=True)
        os.makedirs(output_sparse_dir, exist_ok=True)

        write_images_binary(os.path.join(output_sparse_dir, "images.bin"), split_images)
        
        shutil.copy(os.path.join(unified_model_path, "cameras.bin"), output_sparse_dir)
        shutil.copy(os.path.join(unified_model_path, "points3D.bin"), output_sparse_dir)
        
        ply_path = os.path.join(output_sparse_dir, "points3D.ply")
        cmd_converter = f'{colmap_command} model_converter --input_path "{unified_model_path}" --output_path "{ply_path}" --output_type PLY'
        run_command(cmd_converter)
        
        logging.info(f"已為 {split} 生成最終的 sparse 文件。")

    # --- 步驟 5: 清理 ---
    shutil.rmtree(work_dir)
    logging.info("臨時工作目錄已清理。")
    logging.info(f"===== 所有處理完成！ =====")

if __name__ == '__main__':
    parser = argparse.ArgumentParser(description="全自動執行 COLMAP 重建並分割訓練/測試集。")
    parser.add_argument('--data_path', type=str, required=True, help='包含 train 和 test 的根目錄路徑。')
    parser.add_argument('--colmap_executable', type=str, default="", help='(可選) COLMAP 可執行文件的路徑。')
    args = parser.parse_args()

    run_full_colmap_and_split(args.data_path, args.colmap_executable)