deepfakes/faceswap · error · FaceswapError

Not enough RAM available to sort faces. Try reducing the siz

Error message

Not enough RAM available to sort faces. Try reducing the size of  your dataset. Free RAM: {int(free_ram)}MB. Required RAM: {int(vector_required)}MB

What it means

In identity-based face sorting, faceswap estimates RAM needed for linkage clustering vs. vector clustering. If even the cheaper vector method's requirement exceeds free system RAM, this FaceswapError aborts the sort rather than swapping the machine to death.

Source

Thrown at lib/infer/identity.py:574

        divider = 1024 * 1024  # bytes to MB

        free_ram = psutil.virtual_memory().available / divider
        linkage_required = (((self._num_predictions ** 2) * np_float) / 1.8) / divider
        vector_required = ((self._num_predictions * dims) * np_float) / divider
        logger.debug("free_ram: %sMB, linkage_required: %sMB, vector_required: %sMB",
                     int(free_ram), int(linkage_required), int(vector_required))

        if linkage_required < free_ram:
            logger.verbose("Using linkage method")  # type:ignore[attr-defined]
            retval = False
        elif vector_required < free_ram:
            logger.warning("Not enough RAM to perform linkage clustering. Using vector "
                           "clustering. This will be significantly slower. Free RAM: %sMB. "
                           "Required for linkage method: %sMB",
                           int(free_ram), int(linkage_required))
            retval = True
        else:
            raise FaceswapError("Not enough RAM available to sort faces. Try reducing "
                                f"the size of  your dataset. Free RAM: {int(free_ram)}MB. "
                                f"Required RAM: {int(vector_required)}MB")
        logger.debug(retval)
        return retval

    def _do_linkage(self,
                    predictions: np.ndarray,
                    method: T.Literal["single", "centroid", "median", "ward"]) -> np.ndarray:
        """Use FastCluster to perform vector or standard linkage

        Parameters
        ----------
        predictions
            A stacked matrix of identity predictions of the shape (`N`, `D`) where `N` is the
            number of observations and `D` are the number of dimensions.
        method
            The clustering method to use.

View on GitHub (pinned to f530cb7508)

Solutions

Reduce the dataset size: split faces into subsets and sort each separately.
Free RAM: close other applications, drop caches, increase container/pod memory limit.
Move to a machine with more RAM for the final sort.
Retry after freeing memory — free_ram is measured at runtime.

Example fix

# before
$ python tools.py sort -i /faces -t identity -o /sorted
# FaceswapError: Not enough RAM...

# after (split then sort)
$ split -n l/4 /faces /faces_part_
$ for d in /faces_part_*; do python tools.py sort -i $d -t identity -o ${d}_sorted; done

Defensive patterns

Strategy: validation

Validate before calling

import psutil, math

free_ram_mib = psutil.virtual_memory().available >> 20
est_required_mib = (num_faces ** 2) * 8 >> 20  # vector method ~ N^2 float64
if est_required_mib > free_mib:
    raise SystemExit('dataset too large for in-RAM sort; split it first')

Try / catch

try:
    sort_by_identity(faces)
except FaceswapError as err:
    if 'Not enough RAM' in str(err):
        for chunk in split(faces, n=4):
            sort_by_identity(chunk)
    else:
        raise

Prevention

Estimate N^2 memory cost before sorting large face sets.
Free page cache and close apps before identity sorting.
Give containers a memory limit comfortably above the N^2 estimate.

When it happens

Trigger: Running sort by identity on a very large face set (hundreds of thousands of embeddings) on a machine with insufficient free RAM; free RAM already consumed by caches or other processes.

Common situations: Sorting massive extracted datasets on 8-16GB machines; running sort right after extraction while caches are full; container memory limits lower than host RAM.

Related errors

AI-assisted analysis of deepfakes/faceswap@f530cb7508 (2026-08-15). Data as JSON: /api/errors/2aac6df0237e33ba. Report an issue: GitHub.