Add One-to-All refine helper

One-to-All-Animation/video-generation/infer_refine.sh

@@ -0,0 +1,189 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+WORKDIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+PYTHON_BIN="${ONE_TO_ALL_PYTHON:-python}"
+
+INPUT_PATH="/research/cbim/vast/sf895/code/SignerX-inference-webui/plugins/StableSigner/generated_sign_video/2026-02-14_ref_2.5_pose_1.5/ref_man_vid_HELLO_GOOD_MORNING_20251215_142658_openpose_style_ref_2.5_pose_1.5"
+REF_PATH="/research/cbim/vast/sf895/code/SignerX-inference-webui/plugins/StableSigner/ref_img_dict/man.jpg"
+ALIGN_MODE="pose"
+DO_ALIGN="1"
+FRAME_INTERVAL="1"
+FRAMES_FPS="25"
+REF_CFG="2.5"
+POSE_CFG="1.5"
+CKPT_PATH="${WORKDIR}/../checkpoints/One-to-All-1.3b_2"
+GPU_ID=""
+OUT_H=""
+OUT_W=""
+TMP_VIDEO=""
+OUTPUT_DIR="${WORKDIR}/../../generated_sign_video_refine"
+
+usage() {
+  cat <<'USAGE'
+Usage:
+  infer_refine.sh --input <video_or_frames_dir> --ref <ref_image> [options]
+
+Options:
+  --input PATH          Input video file or a directory of frames.
+  --ref PATH            Reference image path.
+  --fps N               FPS used only when input is a frames directory (default: 25).
+  --frame-interval N    Sample every N frames during preprocessing (default: 1).
+  --align MODE          Alignment mode: ref|pose (default: pose).
+  --no-align            Disable pose alignment in preprocessing.
+  --ref-cfg FLOAT       Image guidance scale (default: 2.5).
+  --pose-cfg FLOAT      Pose guidance scale (default: 1.5).
+  --ckpt PATH           One-to-All 1.3B checkpoint directory.
+  --out-dir PATH        Output directory for generated videos (default: generated_sign_video_refine).
+  --gpu ID              CUDA device id to use (optional).
+  --h N                 Override output height (optional).
+  --w N                 Override output width (optional).
+  -h, --help            Show this help.
+USAGE
+}
+
+while [[ $# -gt 0 ]]; do
+  case "$1" in
+    --input) INPUT_PATH="$2"; shift 2;;
+    --ref) REF_PATH="$2"; shift 2;;
+    --fps) FRAMES_FPS="$2"; shift 2;;
+    --frame-interval) FRAME_INTERVAL="$2"; shift 2;;
+    --align) ALIGN_MODE="$2"; shift 2;;
+    --no-align) DO_ALIGN="0"; shift 1;;
+    --ref-cfg) REF_CFG="$2"; shift 2;;
+    --pose-cfg) POSE_CFG="$2"; shift 2;;
+    --ckpt) CKPT_PATH="$2"; shift 2;;
+    --out-dir) OUTPUT_DIR="$2"; shift 2;;
+    --gpu) GPU_ID="$2"; shift 2;;
+    --h) OUT_H="$2"; shift 2;;
+    --w) OUT_W="$2"; shift 2;;
+    -h|--help) usage; exit 0;;
+    *) echo "Unknown option: $1"; usage; exit 1;;
+  esac
+done
+
+if [[ -z "${INPUT_PATH}" || -z "${REF_PATH}" ]]; then
+  usage
+  exit 1
+fi
+
+if [[ -n "${GPU_ID}" ]]; then
+  export CUDA_VISIBLE_DEVICES="${GPU_ID}"
+fi
+export ONE_TO_ALL_OUTPUT_DIR="${OUTPUT_DIR}"
+
+cleanup() {
+  if [[ -n "${TMP_VIDEO}" && -f "${TMP_VIDEO}" ]]; then
+    rm -f "${TMP_VIDEO}"
+  fi
+}
+trap cleanup EXIT
+
+VIDEO_PATH="${INPUT_PATH}"
+if [[ -d "${INPUT_PATH}" ]]; then
+  pattern=""
+  if ls "${INPUT_PATH}"/*.png >/dev/null 2>&1; then
+    pattern="*.png"
+  elif ls "${INPUT_PATH}"/*.jpg >/dev/null 2>&1; then
+    pattern="*.jpg"
+  elif ls "${INPUT_PATH}"/*.jpeg >/dev/null 2>&1; then
+    pattern="*.jpeg"
+  else
+    echo "No PNG/JPG frames found in ${INPUT_PATH}" >&2
+    exit 1
+  fi
+
+  TMP_VIDEO="$(mktemp -p /tmp one_to_all_input_XXXXXX.mp4)"
+  ffmpeg -y -framerate "${FRAMES_FPS}" -pattern_type glob -i "${INPUT_PATH}/${pattern}" \
+    -c:v libx264 -pix_fmt yuv420p "${TMP_VIDEO}"
+  VIDEO_PATH="${TMP_VIDEO}"
+elif [[ -n "${FRAMES_FPS}" && "${FRAMES_FPS}" != "25" ]]; then
+  echo "Note: --fps is ignored for video inputs; using source video FPS." >&2
+fi
+
+if [[ -f "${VIDEO_PATH}" ]]; then
+  src_fps="$(ffprobe -v error -select_streams v:0 -show_entries stream=avg_frame_rate -of default=nk=1:nw=1 "${VIDEO_PATH}" || true)"
+  if [[ -n "${src_fps}" ]]; then
+    echo "Using source video FPS: ${src_fps}" >&2
+  else
+    echo "Using source video FPS (unable to detect via ffprobe)." >&2
+  fi
+fi
+
+cd "${WORKDIR}"
+
+cache_file="$(mktemp -p /tmp one_to_all_cache_XXXXXX.txt)"
+CACHE_DIR="$("${PYTHON_BIN}" - <<PY
+import os, glob
+from infer_preprocess import process_one
+
+ref_path = r"""${REF_PATH}"""
+video_path = r"""${VIDEO_PATH}"""
+frame_interval = int("${FRAME_INTERVAL}")
+alignmode = r"""${ALIGN_MODE}"""
+do_align = "${DO_ALIGN}" == "1"
+h = r"""${OUT_H}"""
+w = r"""${OUT_W}"""
+h = int(h) if h else None
+w = int(w) if w else None
+
+process_one(
+    reference_path=ref_path,
+    video_path=video_path,
+    frame_interval=frame_interval,
+    do_align=do_align,
+    alignmode=alignmode,
+    h=h,
+    w=w,
+    face_change=True,
+    head_change=True,
+    without_face=False,
+)
+
+refname = os.path.splitext(os.path.basename(ref_path))[0]
+vidname = os.path.splitext(os.path.basename(video_path))[0]
+pattern = os.path.join("..", "input_cache", f"ref_{refname}_driven_{vidname}_align_{alignmode}_*")
+matches = glob.glob(pattern)
+if not matches:
+    raise SystemExit(f"No cache dir found: {pattern}")
+cache_dir = max(matches, key=lambda p: os.path.getmtime(p))
+# Write to a file to avoid stdout contamination by warnings.
+with open(r"""${cache_file}""", "w", encoding="utf-8") as f:
+    f.write(cache_dir)
+PY
+)"
+CACHE_DIR="$(cat "${cache_file}")"
+rm -f "${cache_file}"
+
+"${PYTHON_BIN}" - <<PY
+import os
+import importlib.util
+
+workdir = r"""${WORKDIR}"""
+inference_path = os.path.join(workdir, "inference_1.3b.py")
+spec = importlib.util.spec_from_file_location("inference_1_3b", inference_path)
+mod = importlib.util.module_from_spec(spec)
+spec.loader.exec_module(mod)
+
+cache_dir = r"""${CACHE_DIR}"""
+ckpt_path = r"""${CKPT_PATH}"""
+ref_cfg = float("${REF_CFG}")
+pose_cfg = float("${POSE_CFG}")
+refname = os.path.basename(r"""${REF_PATH}""")
+vidname = os.path.basename(r"""${VIDEO_PATH}""")
+refname = os.path.splitext(refname)[0]
+vidname = os.path.splitext(vidname)[0]
+job_name = f"ref_{refname}_vid_{vidname}"
+
+task_list = [{
+    "cache_dir": cache_dir,
+    "job_name": job_name,
+    "new_h": "",
+    "new_w": "",
+    "ref_cfg": ref_cfg,
+    "pose_cfg": pose_cfg,
+    "prompt": ""
+}]
+
+mod.unified_worker(0, 1, task_list, ckpt_path, use_direct_paths=False)
+PY

One-to-All-Animation/video-generation/inference_1.3b.py

@@ -37,7 +37,7 @@ cache_base_dir="../input_cache"
 # Modified: Use StableSigner3 generated pose videos
 pose_video_dir = "../../generated_pose_video"
 ref_image_dir = "../../ref_img_dict"
-output_base_dir = "../../generated_sign_video"  # 修改：输出到 generated_sign_video
+output_base_dir = os.getenv("ONE_TO_ALL_OUTPUT_DIR", "../../generated_sign_video")
 max_short = 384
 
 MAIN_CHUNK = 81