Deepfake Detection Model Weights
Model checkpoints for 22 completed image and video experiments, uploaded directly from
the source repo's train/ tree (see huggingface_uploader.py in the source repo). No local
staging copy is maintained -- this repo IS the distribution artifact.
Layout
image/<Family>_<model_name>/<run_name>/best.pth
image/<Family>_<model_name>/<run_name>/config.json
image/<Family>_<model_name>/<run_name>/final_summary.json
video/<Family>_<model_name>/<run_name>/... (same shape)
Only the best checkpoint per run (not last.pth, not the duplicate final_*.pth copy that
exists in the source repo). No training logs, no per-sample prediction CSVs, no .md files.
Pulling from a backend
from huggingface_hub import snapshot_download
# everything
local_dir = snapshot_download(repo_id="Anson-Saju-George/deepfake-model-weights")
# just one model (avoids pulling the whole multi-GB tree)
local_dir = snapshot_download(
repo_id="Anson-Saju-George/deepfake-model-weights",
allow_patterns=["image/ConvNeXt_convnext_base/*"],
)