Spaces:
Running
Running
File size: 7,905 Bytes
8a16253 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 | """
Batch QA harness — swap many SOURCE faces onto every gender + location, save the
results, build a contact-sheet grid per location, and write a metrics report.
Put source faces in:
tests/faces/male/*.jpg tests/faces/female/*.jpg
(or pass --faces <dir>). If empty, falls back to external/HF_weights/input/.
Run on the GPU machine:
python scripts/batch_test.py # face-only, all locations (fast)
python scripts/batch_test.py --hair # include HairFastGAN (slow!)
python scripts/batch_test.py --gender Male --limit 5
python scripts/batch_test.py --locations "ICT Department,Library"
Outputs:
tests/output/<gender>/<location>/<source>.jpg
tests/output/<gender>/<location>/_grid.jpg (contact sheet)
tests/output/report.csv (alignment/blend/ΔE/naturalness)
"""
import os, sys, csv, glob, time, argparse
sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
os.environ.setdefault("PYTHONIOENCODING", "utf-8")
try:
from dotenv import load_dotenv; load_dotenv()
except Exception:
pass
import cv2
import numpy as np
from utils.image_io import resize_keep_aspect
from core.detector import detect_faces, _get_insightface
from core.swapper import swap_face_insightface
from core.super_res import restore_faces, _device
from core.head_swap import (full_head_swap, swap_hair, match_skin_to_source,
transfer_glasses)
from core.hair_transfer import transfer_hair
from core.blender import laplacian_blend
from core.segmentor import segment_hair_neck_skin
from core.quality_checker import compute_quality_score, _naturalness_score
from core import supabase_store
ROOT = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
FACES_DIR = os.path.join(ROOT, "tests", "faces")
OUT_DIR = os.path.join(ROOT, "tests", "output")
LOC_DIR = os.path.join(ROOT, "Location")
EXTS = (".jpg", ".jpeg", ".png", ".webp", ".JPG", ".JPEG", ".PNG")
def list_sources(gender, faces_dir):
d = os.path.join(faces_dir, gender.lower())
files = [f for f in sorted(glob.glob(os.path.join(d, "*"))) if f.endswith(EXTS)]
if not files: # fallback sample set
files = sorted(glob.glob(os.path.join(ROOT, "external", "HF_weights", "input", "*.png")))
return files
def list_locations(gender):
"""Locations with a usable target image (Supabase if enabled, else local)."""
if supabase_store.is_enabled():
return [(l["folder"], None) for l in supabase_store.list_locations(gender) if l["has_image"]]
out = []
gdir = os.path.join(LOC_DIR, gender)
if os.path.isdir(gdir):
for name in sorted(os.listdir(gdir)):
folder = os.path.join(gdir, name)
if os.path.isdir(folder):
imgs = [f for f in sorted(os.listdir(folder)) if f.endswith(EXTS)]
if imgs:
out.append((name, os.path.join(folder, imgs[0])))
return out
def load_target(gender, folder, local_path):
if local_path:
return cv2.imread(local_path)
data = supabase_store.get_image_bytes(gender, folder)
if not data:
return None
return cv2.imdecode(np.frombuffer(data, np.uint8), cv2.IMREAD_COLOR)
def run_swap(src, tgt, with_hair):
fs, ft = detect_faces(src), detect_faces(tgt)
if not fs or not ft:
return None, {}
base = tgt
hs = full_head_swap(src, tgt)
if hs is not None:
base = hs
sw = swap_face_insightface(src, base)
sw = restore_faces(sw)
sw = match_skin_to_source(sw, src, fs[0], ft[0], strength=0.75)
if with_hair:
try:
hf = transfer_hair(face_bgr=sw, shape_bgr=src, color_bgr=src)
if hf is not None:
pad = int(max(hf.shape[:2]) * 0.4)
hfp = cv2.copyMakeBorder(hf, pad, pad, pad, pad, cv2.BORDER_CONSTANT, value=(127, 127, 127))
sw = swap_hair(sw, hfp, sw, include_face=False)
except Exception as e:
print(" hair err:", e)
try:
sw = transfer_glasses(sw, src)
fm = segment_hair_neck_skin(sw).get("face_mask")
if fm is not None and fm.max() > 0:
sw = laplacian_blend(sw, tgt, fm, levels=4)
except Exception:
pass
q = compute_quality_score(sw, tgt, None, None)
return sw, q
def grid(images, cols=5, cell=220):
if not images:
return None
rows = (len(images) + cols - 1) // cols
canvas = np.full((rows * cell, cols * cell, 3), 240, np.uint8)
for i, im in enumerate(images):
r, c = divmod(i, cols)
thumb = resize_keep_aspect(im, cell - 8)
h, w = thumb.shape[:2]
y, x = r * cell + (cell - h) // 2, c * cell + (cell - w) // 2
canvas[y:y+h, x:x+w] = thumb
return canvas
def main():
ap = argparse.ArgumentParser()
ap.add_argument("--faces", default=FACES_DIR)
ap.add_argument("--gender", choices=["Male", "Female", "both"], default="both")
ap.add_argument("--locations", default="", help="comma-separated location labels (default: all)")
ap.add_argument("--limit", type=int, default=0, help="max sources per gender (0 = all)")
ap.add_argument("--hair", action="store_true", help="include HairFastGAN (slow)")
args = ap.parse_args()
if _get_insightface() is None:
print("InsightFace not available — aborting."); return 1
print(f"device: {_device()} hair: {args.hair}")
genders = ["Male", "Female"] if args.gender == "both" else [args.gender]
want_locs = {s.strip().lower() for s in args.locations.split(",") if s.strip()}
os.makedirs(OUT_DIR, exist_ok=True)
rows = [("gender", "location", "source", "alignment", "blend", "delta_e", "naturalness", "seconds")]
total = 0
for gender in genders:
sources = list_sources(gender, args.faces)
if args.limit:
sources = sources[:args.limit]
locs = list_locations(gender)
if want_locs:
from core.detector import _get_cascade # noqa
def clean(n): import re; return re.sub(r"^\s*\d+\.\s*", "", n).strip().lower()
locs = [(f, p) for (f, p) in locs if clean(f) in want_locs]
print(f"\n== {gender}: {len(sources)} sources x {len(locs)} locations ==")
for folder, lpath in locs:
tgt = load_target(gender, folder, lpath)
if tgt is None:
print(f" [skip] {folder}: no target"); continue
tgt = resize_keep_aspect(tgt, 1536 if _device() == "cuda" else 1024)
odir = os.path.join(OUT_DIR, gender, folder.replace("/", "_"))
os.makedirs(odir, exist_ok=True)
results = []
for sp in sources:
src = resize_keep_aspect(cv2.imread(sp), 1536 if _device() == "cuda" else 1024)
name = os.path.splitext(os.path.basename(sp))[0]
t = time.time()
sw, q = run_swap(src, tgt, args.hair)
dt = time.time() - t
if sw is None:
print(f" [fail] {folder} <- {name}"); continue
cv2.imwrite(os.path.join(odir, f"{name}.jpg"), sw)
results.append(sw)
rows.append((gender, folder, name, q.get("alignment"), q.get("blend"),
q.get("delta_e"), q.get("naturalness"), round(dt, 1)))
total += 1
print(f" ok {folder} <- {name} ({dt:.1f}s) align={q.get('alignment')} nat={q.get('naturalness')}")
g = grid(results)
if g is not None:
cv2.imwrite(os.path.join(odir, "_grid.jpg"), g)
with open(os.path.join(OUT_DIR, "report.csv"), "w", newline="", encoding="utf-8") as f:
csv.writer(f).writerows(rows)
print(f"\nDone — {total} swaps. Report: tests/output/report.csv")
return 0
if __name__ == "__main__":
sys.exit(main())
|