The place to discuss Faceswap and Deepfakes
https://forum.faceswap.dev/
Hello - I just setup Faceswap for the first time in years so I can get back into it, but both the CLI and the GUI seems to be ignoring my "extract-every-n" setting. Instead of checking every 12th frame, it seems to be checking every frame, producing far too many images.
I tried digging into the code, and it seems the -N option only servers to filter an input file list, not to skip frames when extracting faces from a video.
I recall the -een flag used to skip video frames back when I first started using Faceswap a few years ago, and it seems to now be deprecated in favor of -N, but the new flag doesn't do what the old one did.
I finally found the place in the code to hack to skip frames before they're processed for faces:
Code: Select all
# faceswap/lib/image.py
class ImagesLoader(ImageIO):
...
def _from_video(self):
...
for idx, frame in enumerate(reader):
# HACK GOES HERE
...Now I just need to figure out how to drive that hack using the -N arg.
Is it worth submitting a patch? Is the project still active? (Last commit on 'fs3' branch was seven months ago.)
Here's the full diff I ended up with:
Code: Select all
diff --git a/lib/image.py b/lib/image.py
index 897d8e2..e3a8298 100644
--- a/lib/image.py
+++ b/lib/image.py
@@ -1073,6 +1073,7 @@ class ImagesLoader(ImageIO):
def __init__(self,
path,
+ een,
queue_size=8,
fast_count=True,
skip_list=None,
@@ -1087,6 +1088,7 @@ class ImagesLoader(ImageIO):
self._fps = self._get_fps()
self._count = None
+ self._een = een if een > 1 else None
self._file_list = None
self._get_count_and_filelist(fast_count, count)
@@ -1242,6 +1244,9 @@ class ImagesLoader(ImageIO):
logger.debug("Loading frames from video: '%s'", self.location)
reader = imageio.get_reader(self.location, "ffmpeg")
for idx, frame in enumerate(reader):
+ if self._een and idx % self._een != 0:
+ logger.trace("Skipping frame %s due to een", idx)
+ continue
if idx in self._skip_list:
logger.trace("Skipping frame %s due to skip list", idx)
continue
diff --git a/scripts/extract.py b/scripts/extract.py
index a4912d5..fc0bffa 100644
--- a/scripts/extract.py
+++ b/scripts/extract.py
@@ -496,10 +496,11 @@ class PipelineLoader():
def __init__(self,
path: str | list[str],
extractor: Extractor,
+ een: int,
aligned_filenames: list[str] | None = None) -> None:
logger.debug("Initializing %s: (path: %s, extractor: %s, aligned_filenames: %s)",
self.__class__.__name__, path, extractor, aligned_filenames)
- self._images = ImagesLoader(path, fast_count=True)
+ self._images = ImagesLoader(path, fast_count=True, een=een)
self._extractor = extractor
self._threads: list[MultiThread] = []
self._aligned_filenames = [] if aligned_filenames is None else aligned_filenames
@@ -639,7 +640,7 @@ class _Extract():
self._args.output_dir)
logger.info("Output Directory: %s", self._output_dir)
- self._loader = PipelineLoader(self._args.input_dir, extractor)
+ self._loader = PipelineLoader(self._args.input_dir, extractor, een=self._args.extract_every_n)
self._alignments = Alignments(self._args, True, self._loader.is_video)
self._extractor = extractorI have not tested this, but I believe this issue is a regression because imageio-ffmpeg updated to a new version (as EEN definitely worked in the past, as I used it extensively)
I have now pinned to an older version of imageio-ffmpeg, so you may need to re-install faceswap. Hopefully this will fix the issue, but I would be keen to know either way.