Python | OpenCV 資料蒐集

# download_images.py
from icrawler.builtin import BingImageCrawler

crawler = BingImageCrawler(storage={"root_dir": "dataset/cat"})
crawler.crawl(keyword="cat", max_num=200)

# capture_from_camera.py
import cv2
import os

save_dir = "dataset/custom"
os.makedirs(save_dir, exist_ok=True)

cap = cv2.VideoCapture(0)
count = 0

print("按下空白鍵拍照，按 q 離開")

while True:
    ret, frame = cap.read()
    if not ret:
        break

    cv2.imshow("Capture", frame)
    key = cv2.waitKey(1) & 0xFF

    if key == ord(' '):
        filename = os.path.join(save_dir, f"{count:04d}.jpg")
        cv2.imwrite(filename, frame)
        print(f"已儲存：{filename}")
        count += 1
    elif key == ord('q'):
        break

cap.release()
cv2.destroyAllWindows()

# extract_frames.py
import cv2
import os

video_path = "video.mp4"
save_dir = "dataset/frames"
os.makedirs(save_dir, exist_ok=True)

cap = cv2.VideoCapture(video_path)
frame_interval = 10  # 每隔 10 幀截取一張
count = 0
saved = 0

while True:
    ret, frame = cap.read()
    if not ret:
        break
    if count % frame_interval == 0:
        filename = os.path.join(save_dir, f"{saved:04d}.jpg")
        cv2.imwrite(filename, frame)
        saved += 1
    count += 1

cap.release()
print(f"共截取 {saved} 張影格")

# diagnose_video.py
import cv2
import numpy as np

VIDEO_PATH = "toy_car.mp4"

cap = cv2.VideoCapture(VIDEO_PATH)
fps = cap.get(cv2.CAP_PROP_FPS) or 30.0
step = max(1, int(fps * 0.5))
values = []
frame_idx = 0

while True:
    ret, frame = cap.read()
    if not ret:
        break
    if frame_idx % step == 0:
        gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
        values.append(cv2.Laplacian(gray, cv2.CV_64F).var())
    frame_idx += 1

cap.release()
values.sort()
print(f"取樣幀數：{len(values)}")
print(f"最小值：{min(values):.1f}  最大值：{max(values):.1f}")
print(f"中位數：{np.median(values):.1f}  平均值：{np.mean(values):.1f}")
print(f"建議 BLUR_THRESHOLD：{np.median(values) * 0.5:.1f}")

# collect_from_video.py
import cv2
import os
import numpy as np

# ── 設定 ────────────────────────────────────────────────────────
VIDEO_PATH      = "toy_car.mp4"      # 輸入影片路徑
SAVE_DIR        = "dataset/toy_car"  # 輸出目錄
INTERVAL_SEC    = 0.5                # 每 0.5 秒取一幀
                                     # 手持繞物建議 0.5；轉盤建議 1.0（每秒一幀，一分鐘約 60 張）
BLUR_THRESHOLD  = 21.6               # Laplacian 變異數，低於此值視為模糊幀並跳過
                                     # 依 diagnose_video.py 的建議值填入；手機直拍約 50~200
HASH_THRESHOLD  = 1                  # 感知雜湊漢明距離，小於此值視為重複幀並跳過
                                     # 手持繞物建議 5~8；轉盤固定背景建議設 1（等同關閉重複過濾）

os.makedirs(SAVE_DIR, exist_ok=True)


def is_blurry(frame, threshold):
    gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
    return cv2.Laplacian(gray, cv2.CV_64F).var() < threshold


def phash(frame):
    gray  = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
    small = cv2.resize(gray, (8, 8), interpolation=cv2.INTER_AREA)
    bits  = (small > small.mean()).flatten()
    return "".join("1" if b else "0" for b in bits)


def hamming(a, b):
    return sum(x != y for x, y in zip(a, b))


def collect(video_path, save_dir, interval_sec, blur_thresh, hash_thresh):
    cap = cv2.VideoCapture(video_path)
    if not cap.isOpened():
        print(f"❌ 無法開啟影片：{video_path}")
        return

    fps     = cap.get(cv2.CAP_PROP_FPS) or 30.0
    total_f = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
    step    = max(1, int(fps * interval_sec))

    print(f"影片資訊：{total_f} 幀 / {fps:.1f}fps / {total_f/fps:.1f}s")
    print(f"模糊閾值：{blur_thresh}  |  重複閾值：< {hash_thresh}")
    print("-" * 50)

    saved, skip_blur, skip_dup = 0, 0, 0
    frame_idx = 0
    hash_pool = []

    while True:
        ret, frame = cap.read()
        if not ret:
            break

        if frame_idx % step == 0:
            if is_blurry(frame, blur_thresh):
                skip_blur += 1
            else:
                h = phash(frame)
                if any(hamming(h, ph) < hash_thresh for ph in hash_pool):
                    skip_dup += 1
                else:
                    fname = os.path.join(save_dir, f"{saved:05d}.jpg")
                    cv2.imwrite(fname, frame, [cv2.IMWRITE_JPEG_QUALITY, 95])
                    hash_pool.append(h)
                    saved += 1

        frame_idx += 1

        if frame_idx % 100 == 0:
            pct = frame_idx / total_f * 100
            print(f"\r進度：{pct:5.1f}%  已儲存：{saved}  "
                  f"跳過(模糊)：{skip_blur}  跳過(重複)：{skip_dup}",
                  end="", flush=True)

    cap.release()
    print(f"\n{'=' * 50}")
    print(f"完成！輸出目錄：{save_dir}")
    print(f"  ✅ 儲存：{saved} 張")
    print(f"  🌀 跳過（模糊）：{skip_blur} 張")
    print(f"  📋 跳過（重複）：{skip_dup} 張")


if __name__ == "__main__":
    collect(VIDEO_PATH, SAVE_DIR, INTERVAL_SEC, BLUR_THRESHOLD, HASH_THRESHOLD)

# check_images.py
import cv2
import os

def check_images(directory):
    broken = []
    for root, _, files in os.walk(directory):
        for f in files:
            if f.lower().endswith(('.jpg', '.jpeg', '.png')):
                path = os.path.join(root, f)
                if cv2.imread(path) is None:
                    broken.append(path)
    return broken

broken_files = check_images("dataset")
print(f"損壞圖片數量：{len(broken_files)}")
for f in broken_files:
    print(f)