首次提交
This commit is contained in:
480
tool/VideoProcessing/multithread_algo.py
Normal file
480
tool/VideoProcessing/multithread_algo.py
Normal file
@@ -0,0 +1,480 @@
|
||||
from tool.log.log_manager import logger
|
||||
from config.config_utils import config, get_ffmpeg, get_ffprobe
|
||||
import shutil
|
||||
import subprocess
|
||||
import json
|
||||
import cv2
|
||||
import os
|
||||
import itertools
|
||||
import threading
|
||||
import time
|
||||
|
||||
|
||||
def extract_frames(
|
||||
video_path,
|
||||
video_name,
|
||||
start_frame,
|
||||
end_frame,
|
||||
output_dir,
|
||||
frames_to_extract,
|
||||
frame_prefix=""
|
||||
):
|
||||
"""
|
||||
进行抽帧
|
||||
:param video_path: 视频地址
|
||||
:param video_name: 视频名称
|
||||
:param start_frame: 开始的帧
|
||||
:param end_frame: 结束的帧
|
||||
:param output_dir: 输出地址
|
||||
:param frames_to_extract: 帧时间列表
|
||||
:param frame_prefix:
|
||||
:return: 抽取的帧数
|
||||
"""
|
||||
cap = cv2.VideoCapture(video_path)
|
||||
try:
|
||||
extracted_count = 0
|
||||
for current_time in frames_to_extract:
|
||||
# 只抽取在当前线程范围内且预定义的帧
|
||||
if start_frame <= current_time < end_frame:
|
||||
current_ms = 1000 * current_time
|
||||
cap.set(cv2.CAP_PROP_POS_MSEC, current_ms)
|
||||
ret, frame = cap.read()
|
||||
if not ret:
|
||||
print(f"{current_time}无ret")
|
||||
break
|
||||
frame_filename = os.path.join(
|
||||
output_dir,
|
||||
f"{video_name}{frame_prefix}_{3 * int(current_time) + int((current_time - int(current_time) + 0.1) / 0.33)}.jpg"
|
||||
)
|
||||
cv2.imwrite(frame_filename, frame)
|
||||
extracted_count += 1
|
||||
logger.log(f"线程 {start_frame}-{end_frame}: 抽取了{extracted_count}帧", program="split_frame")
|
||||
return extracted_count
|
||||
except Exception as e:
|
||||
logger.log(f"线程 {start_frame}-{end_frame} 抽帧异常: {e}", level="error", program="split_frame")
|
||||
raise
|
||||
finally:
|
||||
cap.release()
|
||||
|
||||
|
||||
def _run_extract_frames(result_container, index, *args, **kwargs):
|
||||
"""包装函数，捕获子线程异常并存储到共享容器"""
|
||||
try:
|
||||
result_container[index] = extract_frames(*args, **kwargs)
|
||||
except Exception as e:
|
||||
result_container[index] = e
|
||||
|
||||
|
||||
def split_video_get_frame_images(
|
||||
video_path,
|
||||
video_name,
|
||||
num_segments,
|
||||
output_dir,
|
||||
interval_seconds=1.0
|
||||
):
|
||||
"""
|
||||
拆分视频，获取视频的帧图片
|
||||
:param video_path:
|
||||
:param video_name:
|
||||
:param num_segments:
|
||||
:param output_dir:
|
||||
:param interval_seconds:
|
||||
:return:
|
||||
"""
|
||||
cap = cv2.VideoCapture(video_path)
|
||||
total_frames = cap.get(cv2.CAP_PROP_FRAME_COUNT)
|
||||
cap_fps = cap.get(cv2.CAP_PROP_FPS)
|
||||
print("total frames:", total_frames)
|
||||
print("fps:", cap_fps)
|
||||
video_duration = total_frames / cap_fps
|
||||
cap.release()
|
||||
frames_to_extract = []
|
||||
if interval_seconds == 0.33:
|
||||
total_seconds = int(video_duration)
|
||||
for seconds in range(total_seconds + 1):
|
||||
for i in range(3):
|
||||
target_time = seconds + i * 0.33
|
||||
if target_time < total_frames:
|
||||
frames_to_extract.append(target_time)
|
||||
# 去重排序
|
||||
frames_to_extract = sorted(set(frames_to_extract))
|
||||
else:
|
||||
t = 0.0
|
||||
while t < video_duration:
|
||||
frames_to_extract.append(t)
|
||||
t += interval_seconds
|
||||
frames_to_extract = sorted(set(frames_to_extract))
|
||||
frames_to_extract = [tt for tt in frames_to_extract if tt < video_duration]
|
||||
expected_frames = len(frames_to_extract)
|
||||
logger.log(f"【视频统计】 视频时长：{video_duration:.2f}秒, 统一计算抽帧位置：{expected_frames}", program="split_frame")
|
||||
|
||||
# 分段处理
|
||||
frames_per_segment = video_duration // num_segments
|
||||
threads = []
|
||||
thread_results = {}
|
||||
|
||||
for i in range(num_segments):
|
||||
start_frame = frames_per_segment * i
|
||||
end_frame = (i + 1) * frames_per_segment if i < num_segments - 1 else video_duration
|
||||
thread = threading.Thread(
|
||||
target=_run_extract_frames,
|
||||
args=(thread_results, i, video_path, video_name, start_frame, end_frame, output_dir, frames_to_extract, "_first")
|
||||
)
|
||||
threads.append(thread)
|
||||
thread.start()
|
||||
|
||||
for thread in threads:
|
||||
thread.join()
|
||||
|
||||
# 检查子线程异常
|
||||
for i, result in thread_results.items():
|
||||
if isinstance(result, Exception):
|
||||
raise RuntimeError(f"线程 {i} 抽帧失败: {result}") from result
|
||||
|
||||
return expected_frames, video_duration
|
||||
|
||||
|
||||
def get_video_codec(video_path: str) -> str:
|
||||
"""通过 ffprobe 获取视频编码格式"""
|
||||
cmd = [
|
||||
get_ffprobe(), '-v', 'error', '-select_streams', 'v:0',
|
||||
'-show_entries', 'stream=codec_name',
|
||||
'-of', 'default=noprint_wrappers=1:nokey=1', video_path
|
||||
]
|
||||
result = subprocess.run(cmd, capture_output=True, text=True)
|
||||
return result.stdout.strip().lower()
|
||||
|
||||
|
||||
# ==================== GPU 硬解检测（结果缓存，只查一次）====================
|
||||
_gpu_lock = threading.Lock()
|
||||
_gpu_checked = False
|
||||
_gpu_available = False
|
||||
_gpu_count = 0 # 检测到的 GPU 张数（多卡轮询用）
|
||||
_gpu_rr = itertools.count() # 多卡轮询计数器（next() 在 CPython 下原子）
|
||||
# GPU 无法解码的编码：首次失败后记下，后续该编码直接走 CPU。
|
||||
# 假设多卡同型号（同样支持/不支持某编码），故按编码全局标记。
|
||||
_gpu_bad_codecs = set()
|
||||
|
||||
|
||||
def gpu_decode_available() -> bool:
|
||||
"""检测本机是否有可用的 NVIDIA GPU 硬解（nvidia-smi 可用 + ffmpeg 支持 cuda），
|
||||
并数出 GPU 张数。结果缓存，整个进程只真正检测一次。"""
|
||||
global _gpu_checked, _gpu_available, _gpu_count
|
||||
if _gpu_checked:
|
||||
return _gpu_available
|
||||
with _gpu_lock:
|
||||
if _gpu_checked:
|
||||
return _gpu_available
|
||||
avail = False
|
||||
count = 0
|
||||
try:
|
||||
r = subprocess.run(['nvidia-smi', '-L'], capture_output=True, text=True, timeout=10)
|
||||
if r.returncode == 0:
|
||||
count = sum(1 for ln in r.stdout.splitlines() if ln.strip().startswith('GPU '))
|
||||
r2 = subprocess.run([get_ffmpeg(), '-hide_banner', '-hwaccels'],
|
||||
capture_output=True, text=True, timeout=10)
|
||||
avail = count > 0 and 'cuda' in (r2.stdout or '').lower()
|
||||
except (FileNotFoundError, OSError, subprocess.TimeoutExpired):
|
||||
avail, count = False, 0
|
||||
_gpu_available = avail
|
||||
_gpu_count = count
|
||||
_gpu_checked = True
|
||||
logger.log(
|
||||
f"GPU 硬解检测: {'可用，抽帧优先用 GPU(cuda)' if avail else '不可用，使用 CPU 软解'}"
|
||||
f"（检测到 {count} 张卡）", program="split_frame")
|
||||
return _gpu_available
|
||||
|
||||
|
||||
# —— 自动探测空闲 GPU（避开被 co-located OCR 服务占用的卡）——
|
||||
_gpu_free_cache = {"ts": 0.0, "free": None} # free=空闲卡索引集合; None=查不到(让上层回退)
|
||||
_gpu_free_lock = threading.Lock()
|
||||
|
||||
|
||||
def _query_gpu_mem_used() -> dict:
|
||||
"""nvidia-smi 查每张卡【已用显存(MB)】, 返回 {index: mem_mb}; 失败返回 {}。"""
|
||||
try:
|
||||
r = subprocess.run(
|
||||
['nvidia-smi', '--query-gpu=index,memory.used',
|
||||
'--format=csv,noheader,nounits'],
|
||||
capture_output=True, text=True, timeout=5)
|
||||
out = {}
|
||||
for ln in r.stdout.splitlines():
|
||||
parts = [p.strip() for p in ln.split(',')]
|
||||
if len(parts) >= 2 and parts[0].isdigit():
|
||||
try:
|
||||
out[int(parts[0])] = float(parts[1])
|
||||
except ValueError:
|
||||
pass
|
||||
return out
|
||||
except Exception:
|
||||
return {}
|
||||
|
||||
|
||||
def _get_free_gpus(busy_mem_mb: float, ttl: float = 20.0):
|
||||
"""返回【空闲】(已用显存 < busy_mem_mb)的 GPU 索引集合; 带 TTL 缓存, 不频繁调 nvidia-smi。
|
||||
返回 None 表示查不到(nvidia-smi 失败), 让上层回退到原有逻辑。"""
|
||||
now = time.time()
|
||||
with _gpu_free_lock:
|
||||
if _gpu_free_cache["free"] is not None and (now - _gpu_free_cache["ts"]) < ttl:
|
||||
return _gpu_free_cache["free"]
|
||||
mem = _query_gpu_mem_used()
|
||||
free = {i for i, m in mem.items() if m < busy_mem_mb} if mem else None
|
||||
_gpu_free_cache["free"] = free
|
||||
_gpu_free_cache["ts"] = now
|
||||
return free
|
||||
|
||||
|
||||
def _next_gpu_device() -> int:
|
||||
"""多卡轮询取下一张 GPU 索引；单卡或未知时返回 0。
|
||||
- gpu_reserve_last_for_whisper: 留最后一张给 whisper（抽帧只用前 N-1 张）
|
||||
- gpu_skip_indices: 手动【跳过】这些卡
|
||||
- gpu_auto_select: 自动探测, 只往【空闲卡】(显存占用 < gpu_busy_mem_mb)上分配,
|
||||
自动避开被 OCR 服务占着、抽帧慢25倍的卡。查不到/全忙时回退正常轮询。"""
|
||||
if _gpu_count <= 1:
|
||||
return 0
|
||||
n = _gpu_count
|
||||
skip = set()
|
||||
auto = False
|
||||
busy_mem = 4096.0
|
||||
try:
|
||||
cfg = config.get_splitframe_config()
|
||||
if cfg.get('gpu_reserve_last_for_whisper', True):
|
||||
n = _gpu_count - 1 # 留最后一张给 whisper
|
||||
for i in (cfg.get('gpu_skip_indices') or []):
|
||||
try:
|
||||
skip.add(int(i))
|
||||
except (TypeError, ValueError):
|
||||
pass
|
||||
auto = bool(cfg.get('gpu_auto_select', False))
|
||||
busy_mem = float(cfg.get('gpu_busy_mem_mb', 4096))
|
||||
except Exception:
|
||||
pass
|
||||
# 候选 = 前 n 张里去掉手动跳过的卡
|
||||
candidates = [i for i in range(max(1, n)) if i not in skip]
|
||||
# 自动探测: 进一步只保留空闲卡(若探测到至少一张空闲); 全忙或查不到则维持原候选
|
||||
if auto:
|
||||
free = _get_free_gpus(busy_mem)
|
||||
if free:
|
||||
free_cands = [i for i in candidates if i in free]
|
||||
if free_cands:
|
||||
candidates = free_cands
|
||||
if not candidates:
|
||||
candidates = list(range(max(1, n))) # 兜底, 不至于不抽帧
|
||||
return candidates[next(_gpu_rr) % len(candidates)]
|
||||
|
||||
|
||||
def _clean_temp(temp_pattern: str):
|
||||
"""清理某次抽帧产生的临时帧（GPU 失败回退前用）。"""
|
||||
d = os.path.dirname(temp_pattern)
|
||||
prefix = os.path.basename(temp_pattern).split('%')[0]
|
||||
try:
|
||||
for f in os.listdir(d):
|
||||
if f.startswith(prefix):
|
||||
try:
|
||||
os.remove(os.path.join(d, f))
|
||||
except OSError:
|
||||
pass
|
||||
except OSError:
|
||||
pass
|
||||
|
||||
|
||||
def _resolve_ffmpeg_threads(raw) -> int:
|
||||
"""splitframe.ffmpeg_threads 支持字面数字，或 "auto"：按 CPU核数/并发进程数 自动摊分，
|
||||
避免 30 台核数不一样的机器都得手改这个数。并发进程数从环境变量 OCR_MAX_WORKERS 读
|
||||
(main.py/run_worker.py 启动时按实际生效的 --max-workers 写入，进程池 forkserver 子进程会继承)；
|
||||
读不到(如手动 import 调试)按 4 估算。0/未配置 = 不限制(交给 ffmpeg 默认行为)。"""
|
||||
if isinstance(raw, str) and raw.strip().lower() == 'auto':
|
||||
cpu = os.cpu_count() or 4
|
||||
try:
|
||||
workers = int(os.environ.get('OCR_MAX_WORKERS', '0') or 0)
|
||||
except ValueError:
|
||||
workers = 0
|
||||
workers = workers if workers > 0 else 4
|
||||
return max(1, cpu // workers)
|
||||
try:
|
||||
return int(raw)
|
||||
except (TypeError, ValueError):
|
||||
return 0
|
||||
|
||||
|
||||
def _run_ffmpeg_extract(video_path: str, fps: float, temp_pattern: str, codec: str) -> str:
|
||||
"""执行抽帧：有 GPU 且该编码可硬解则用 GPU，失败/不支持自动回退 CPU。返回 'gpu' 或 'cpu'。"""
|
||||
cfg = config.get_splitframe_config()
|
||||
use_gpu = cfg.get('use_gpu', True)
|
||||
# 明确不让 GPU 解的编码(如 V100 解不了 AV1 → 配 ["av1"] 直接走 CPU，省掉一次失败探测)
|
||||
skip_codecs = {str(c).lower() for c in (cfg.get('gpu_skip_codecs') or [])}
|
||||
ffmpeg_threads = _resolve_ffmpeg_threads(cfg.get('ffmpeg_threads', 0))
|
||||
|
||||
# 字幕固定在下半屏时，抽帧阶段就裁掉上半部，帧变小 → 下游每帧处理(merge读帧/CV/OCR)成本按比例下降。
|
||||
# crop_bottom_ratio=保留底部比例(如 0.55=保留下 55%)；null/缺省=不裁(全帧)。
|
||||
vf = f'fps={fps:.6f}'
|
||||
crop_r = cfg.get('crop_bottom_ratio')
|
||||
try:
|
||||
crop_r = float(crop_r) if crop_r is not None else None
|
||||
except (TypeError, ValueError):
|
||||
crop_r = None
|
||||
if crop_r is not None and 0.0 < crop_r < 1.0:
|
||||
vf = f'{vf},crop=iw:ih*{crop_r:.4f}:0:ih*{1.0 - crop_r:.4f}'
|
||||
out_opts = ['-vf', vf, '-q:v', '2', '-y', temp_pattern, '-loglevel', 'error']
|
||||
|
||||
# ① GPU 硬解：支持该编码的卡上用 GPU 解码（多卡轮询分配到不同卡）。
|
||||
# 解码在 GPU、自动下载到内存，fps/JPEG 编码在 CPU。
|
||||
# 跳过条件：配置黑名单(gpu_skip_codecs，如 av1) 或 之前解失败过的编码(_gpu_bad_codecs)。
|
||||
# 其他能解的编码（如 H264/HEVC）走 GPU；某编码首次解失败也会记下、之后转 CPU。
|
||||
# ffmpeg 抽帧超时(秒): 防止坏视频/网络读卡死时 worker 永久阻塞。默认1小时,够超长视频抽帧。
|
||||
try:
|
||||
ff_timeout = int(cfg.get('ffmpeg_timeout', 3600))
|
||||
except (TypeError, ValueError):
|
||||
ff_timeout = 3600
|
||||
|
||||
if (use_gpu and codec not in skip_codecs and codec not in _gpu_bad_codecs
|
||||
and gpu_decode_available()):
|
||||
dev = _next_gpu_device()
|
||||
gpu_cmd = [get_ffmpeg()]
|
||||
if ffmpeg_threads > 0:
|
||||
gpu_cmd += ['-threads', str(ffmpeg_threads)]
|
||||
gpu_cmd += ['-hwaccel', 'cuda', '-hwaccel_device', str(dev),
|
||||
'-i', video_path] + out_opts
|
||||
try:
|
||||
subprocess.run(gpu_cmd, check=True, capture_output=True, timeout=ff_timeout)
|
||||
return f'gpu{dev}'
|
||||
except subprocess.TimeoutExpired:
|
||||
logger.log(f"GPU 抽帧超时(>{ff_timeout}s)，回退 CPU: {os.path.basename(video_path)}",
|
||||
level="warning", program="split_frame")
|
||||
_clean_temp(temp_pattern)
|
||||
except subprocess.CalledProcessError as e:
|
||||
with _gpu_lock:
|
||||
_gpu_bad_codecs.add(codec)
|
||||
err = (e.stderr or b'')
|
||||
err = err.decode(errors='ignore') if isinstance(err, bytes) else str(err)
|
||||
logger.log(f"GPU 硬解失败(编码 {codec or '未知'})，该编码后续改用 CPU: {err[:200]}",
|
||||
level="warning", program="split_frame")
|
||||
_clean_temp(temp_pattern) # 清掉半成品再回退
|
||||
|
||||
# ② CPU 软解（限解码线程，避免多视频并发互抢）
|
||||
cmd = [get_ffmpeg()]
|
||||
if ffmpeg_threads > 0:
|
||||
cmd += ['-threads', str(ffmpeg_threads)]
|
||||
cmd += ['-i', video_path] + out_opts
|
||||
# 捕获 stderr：失败时把 ffmpeg 真实报错带出来(否则只看到 exit 255 不知道原因)
|
||||
try:
|
||||
r = subprocess.run(cmd, capture_output=True, text=True, timeout=ff_timeout)
|
||||
except subprocess.TimeoutExpired:
|
||||
_clean_temp(temp_pattern)
|
||||
raise RuntimeError(f"ffmpeg(cpu) 抽帧超时(>{ff_timeout}s): {os.path.basename(video_path)}")
|
||||
if r.returncode != 0:
|
||||
raise RuntimeError(
|
||||
f"ffmpeg(cpu) 退出码 {r.returncode}: {(r.stderr or '').strip()[-600:]}")
|
||||
return 'cpu'
|
||||
|
||||
|
||||
def split_frame_ffmpeg(video_path: str, video_name: str, output_dir: str,
|
||||
interval_seconds: float) -> tuple:
|
||||
"""
|
||||
使用 FFmpeg 提取帧（有 GPU 则硬解加速，否则软解；用于 AV1 等 OpenCV 无法解码的格式）。
|
||||
输出文件命名与 OpenCV 方案保持一致。
|
||||
"""
|
||||
# 获取视频时长与编码格式
|
||||
probe = subprocess.run([
|
||||
get_ffprobe(), '-v', 'error', '-select_streams', 'v:0',
|
||||
'-show_entries', 'stream=duration,codec_name',
|
||||
'-of', 'json', video_path
|
||||
], capture_output=True, text=True)
|
||||
# ffprobe 失败(返回码非0或无输出)时给清楚的错，避免后面 json.loads("") 报看不懂的
|
||||
# "Expecting value"。常见原因：bin/ffprobe 是 GPU 版但本机 glibc 太老(<2.28)跑不起来。
|
||||
if probe.returncode != 0 or not (probe.stdout or "").strip():
|
||||
raise RuntimeError(
|
||||
f"ffprobe 执行失败(返回码 {probe.returncode})；"
|
||||
f"常见原因：bin/ffprobe 与本机不兼容(如 GPU 版 ffmpeg 需 glibc>=2.28)。"
|
||||
f"stderr: {(probe.stderr or '').strip()[:300]}")
|
||||
info = json.loads(probe.stdout)
|
||||
stream0 = info['streams'][0] if info.get('streams') else {}
|
||||
video_duration = float(stream0.get('duration', 0) or 0)
|
||||
codec = (stream0.get('codec_name') or '').lower()
|
||||
# 流时长缺失(某些封装(尤其AV1/部分mp4)只在容器层有duration, 流里没有)时兜底查容器时长——
|
||||
# 不然 video_duration=0 会导致 frames_to_extract 算成空列表(下面 for 循环整个不执行), 但
|
||||
# 实际 ffmpeg 硬解还是会正常跑完整个视频(耗时不受这个影响), 抽完的帧因为"期望帧数=0"在
|
||||
# 后面的重命名循环里被当成多余的全部删掉 —— 表现为"抽帧完成(0帧, 耗时正常, 视频0s)",
|
||||
# 花了几分钟GPU时间白干。跟 tool/process.py 的 _probe_duration() 是同一个坑、同一个修法。
|
||||
if video_duration <= 0:
|
||||
try:
|
||||
fmt_probe = subprocess.run([
|
||||
get_ffprobe(), '-v', 'error',
|
||||
'-show_entries', 'format=duration', '-of', 'json', video_path
|
||||
], capture_output=True, text=True)
|
||||
fmt_info = json.loads(fmt_probe.stdout or '{}')
|
||||
video_duration = float((fmt_info.get('format') or {}).get('duration', 0) or 0)
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
# 计算目标时间点（与 OpenCV 方案一致）
|
||||
frames_to_extract = []
|
||||
if interval_seconds == 0.33:
|
||||
for sec in range(int(video_duration) + 1):
|
||||
for i in range(3):
|
||||
t = sec + i * 0.33
|
||||
if t < video_duration:
|
||||
frames_to_extract.append(t)
|
||||
else:
|
||||
t = 0.0
|
||||
while t < video_duration:
|
||||
frames_to_extract.append(t)
|
||||
t += interval_seconds
|
||||
frames_to_extract = sorted(set(frames_to_extract))
|
||||
expected = len(frames_to_extract)
|
||||
|
||||
# 一次性用 FFmpeg 提取所有帧（fps 滤镜）
|
||||
# 注意：interval 0.33 是"每秒3帧"的约定值，真正频率应为 3.0。
|
||||
# 若用 1/0.33=3.0303，会比真正 3fps 快约1%，导致第N帧实际抽自 N×0.33 秒、
|
||||
# 而 SRT 按 N/3≈N×0.3333 秒标注，产生随帧号累积的时间漂移（18分钟处可差~10秒）。
|
||||
fps = 3.0 if interval_seconds == 0.33 else 1.0 / interval_seconds
|
||||
temp_pattern = os.path.join(output_dir, f"__tmp_{video_name}_%08d.jpg")
|
||||
|
||||
# 有 GPU 用 GPU 硬解，否则 CPU 软解（CPU 路径按 ffmpeg_threads 限解码线程）
|
||||
backend = _run_ffmpeg_extract(video_path, fps, temp_pattern, codec)
|
||||
|
||||
# 按期望时间点重命名
|
||||
temp_files = sorted(
|
||||
f for f in os.listdir(output_dir) if f.startswith(f"__tmp_{video_name}_")
|
||||
)
|
||||
extracted = 0
|
||||
for i, tmp in enumerate(temp_files):
|
||||
if i >= len(frames_to_extract):
|
||||
os.remove(os.path.join(output_dir, tmp))
|
||||
continue
|
||||
ct = frames_to_extract[i]
|
||||
frame_idx = 3 * int(ct) + int((ct - int(ct) + 0.1) / 0.33)
|
||||
new_name = os.path.join(output_dir, f"{video_name}_first_{frame_idx}.jpg")
|
||||
os.rename(os.path.join(output_dir, tmp), new_name)
|
||||
extracted += 1
|
||||
|
||||
logger.log(f"FFmpeg 抽帧完成({backend}): 应抽{expected}帧, 实际{extracted}帧", program="split_frame")
|
||||
return expected, video_duration, backend
|
||||
|
||||
|
||||
def split_frame(video_path, extract_interval=0.33):
|
||||
"""
|
||||
进行视频分割
|
||||
:param video_path: 视频本地位置
|
||||
:param extract_interval: 分割间隔，默认0.33间隔，1秒3张图
|
||||
:return:
|
||||
"""
|
||||
logger.start_program("split_frame")
|
||||
video_name = os.path.basename(video_path).partition(".")[0]
|
||||
to_save_dir = os.path.join(os.getcwd(), "split", video_name)
|
||||
os.makedirs(to_save_dir, exist_ok=True)
|
||||
|
||||
logger.log(f"【 正在抽帧处理 】{video_name}, 抽帧间隔：{extract_interval}秒", program="split_frame")
|
||||
start = time.time()
|
||||
|
||||
# 统一使用 FFmpeg 单次顺序解码（比 OpenCV 逐帧 seek 快 3-5 倍）
|
||||
expected_frames, video_duration, backend = split_frame_ffmpeg(
|
||||
video_path, video_name, to_save_dir, extract_interval)
|
||||
|
||||
end = time.time()
|
||||
|
||||
actual_frames = len([f for f in os.listdir(to_save_dir) if f.endswith('.jpg')])
|
||||
logger.log(f"【抽帧完成】: 应抽{expected_frames}帧, 实际抽取{actual_frames}帧, 耗时{end - start:.2f}秒", program="split_frame")
|
||||
logger.end_program("split_frame")
|
||||
return video_duration, expected_frames, actual_frames, backend
|
||||
|
||||
Reference in New Issue
Block a user