def downscale_frames(self,
subdir,
max_size,
ext,
align=16,
full_subdir="color_full"):
full_dir = pjoin(self.path, full_subdir)
down_dir = pjoin(self.path, subdir)
mkdir_ifnotexists(down_dir)
if self.check_frames(down_dir, ext):
# Frames are already extracted and checked OK.
return
for i in range(self.frame_count):
full_file = "%s/frame_%06d.png" % (full_dir, i)
down_file = ("%s/frame_%06d." + ext) % (down_dir, i)
suppress_messages = (i > 0)
image = image_io.load_image(full_file,
max_size=max_size,
align=align,
suppress_messages=suppress_messages)
image = image[..., ::-1] # Channel swizzle
if ext == "raw":
image_io.save_raw_float32_image(down_file, image)
else:
cv2.imwrite(down_file, image * 255)
self.check_frames(down_dir, ext)
def extract_pts(self):
if self.check_extracted_pts():
# frames.txt exists and checked OK.
return
if not os.path.exists(self.video_file):
sys.exit("ERROR: input video file '%s' not found.",
self.video_file) # 如果视频文件没有找到报错
# Get width and height
tmp_file = tempfile.mktemp(".png")
cmd = "%s -i %s -vframes 1 %s" % (ffmpeg, self.video_file, tmp_file
) # 此命令的作用为输出视频第1帧
print(cmd)
res = os.popen(
cmd).read() # 执行命令行 并获得屏幕输出 ffmpeg -i 视频文件路径 -vframes 1 临时文件
image = image_io.load_image(tmp_file)
height = image.shape[0] # 获得视频的高
width = image.shape[1] # 获得视频的宽度
os.remove(tmp_file)
if os.path.exists(tmp_file): # 检查临时文件是否存在
sys.exit("ERROR: unable to remove '%s'" % tmp_file) # 报告无法删除临时文件
# Get PTS
def parse_line(line, token):
if line[:len(token)] != token:
sys.exit("ERROR: record is malformed, expected to find '%s'." %
token)
return line[len(token):]
ffprobe_cmd = "%s %s -select_streams v:0 -show_frames" % (
ffprobe,
self.video_file,
) # ffprobe 视频地址 -select_streams v:0 -show_frames
cmd = ffprobe_cmd + " | grep pkt_pts_time"
print(cmd)
res = os.popen(cmd).read()
pts = []
for line in res.splitlines():
pts.append(parse_line(
line, "pkt_pts_time=")) # 获得PTS（Presentation Time Stamp）
# https: // zhuanlan.zhihu.com / p / 34162672
self.frame_count = len(pts)
print("%d frames detected." % self.frame_count)
pts_file = "%s/frames.txt" % self.path
with open(pts_file, "w") as file: # 将得到的pts写入文件
file.write("%d\n" % len(pts))
file.write("%s\n" % width)
file.write("%s\n" % height)
for t in pts:
file.write("%s\n" % t)
self.check_extracted_pts() #检查frames.txt 文件是否正确
def extract_pts(self):
if self.check_extracted_pts():
# frames.txt exists and checked OK.
return
# 检查是否存在文件
if not os.path.exists(self.video_file):
sys.exit("ERROR: input video file '%s' not found.",
self.video_file)
# Get width and height
tmp_file = tempfile.mktemp(".png")
cmd = "%s -i %s -vframes 1 %s" % (ffmpeg, self.video_file, tmp_file)
print(cmd)
res = os.popen(cmd).read()
image = image_io.load_image(tmp_file)
height = image.shape[0]
width = image.shape[1]
os.remove(tmp_file) # 删除中间文件
if os.path.exists(tmp_file):
sys.exit("ERROR: unable to remove '%s'" % tmp_file)
# Get PTS
def parse_line(line, token):
if line[:len(token)] != token:
sys.exit("ERROR: record is malformed, expected to find '%s'." %
token)
return line[len(token):]
# 查询视频帧以及关键时间点
ffprobe_cmd = "%s %s -select_streams v:0 -show_frames" % (
ffprobe,
self.video_file,
)
cmd = ffprobe_cmd + " | grep pkt_pts_time"
print(cmd)
res = os.popen(cmd).read()
pts = []
for line in res.splitlines():
pts.append(parse_line(line, "pkt_pts_time="))
self.frame_count = len(pts)
print("%d frames detected." % self.frame_count)
# 设置帧输出文件
pts_file = "%s/frames.txt" % self.path
with open(pts_file, "w") as file:
file.write("%d\n" % len(pts))
file.write("%s\n" % width)
file.write("%s\n" % height)
for t in pts:
file.write("%s\n" % t)
# 再次确认文件
self.check_extracted_pts()