Spaces:
Sleeping
Sleeping
Matan Kriel commited on
Commit Β·
2277fc2
1
Parent(s): 72e2859
added new parquet file
Browse files
app.py
CHANGED
|
@@ -27,6 +27,8 @@ else:
|
|
| 27 |
if DB_PATH:
|
| 28 |
print(f"π Loaded Knowledge Base: {DB_PATH}")
|
| 29 |
df_db = pd.read_parquet(DB_PATH)
|
|
|
|
|
|
|
| 30 |
# Convert embedding column to a clean numpy matrix for fast math
|
| 31 |
DB_VECTORS = np.stack(df_db['embedding'].values)
|
| 32 |
|
|
@@ -34,6 +36,20 @@ if DB_PATH:
|
|
| 34 |
# e.g. "famous_faces_GhostFaceNet.parquet" -> "GhostFaceNet"
|
| 35 |
MODEL_NAME = DB_PATH.split("_")[-1].replace(".parquet", "")
|
| 36 |
print(f"βοΈ Model configured: {MODEL_NAME}")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 37 |
else:
|
| 38 |
print("β CRITICAL: No Parquet file found! Please run the Model Battle step.")
|
| 39 |
DB_VECTORS = None
|
|
@@ -41,49 +57,87 @@ else:
|
|
| 41 |
|
| 42 |
# --- 2. Helper function to load images with fallbacks ---
|
| 43 |
def load_image_with_fallback(image_path, name):
|
| 44 |
-
"""
|
| 45 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 46 |
if os.path.exists(image_path):
|
| 47 |
try:
|
| 48 |
-
return Image.open(image_path)
|
| 49 |
-
except:
|
| 50 |
-
|
|
|
|
|
|
|
| 51 |
|
| 52 |
-
# Try just the filename in current directory
|
| 53 |
filename = os.path.basename(image_path)
|
| 54 |
if os.path.exists(filename):
|
| 55 |
try:
|
| 56 |
-
return Image.open(filename)
|
| 57 |
-
except:
|
| 58 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 59 |
|
| 60 |
-
# Try in my_dataset directory
|
| 61 |
-
dataset_path = os.path.join("my_dataset", filename)
|
| 62 |
if os.path.exists(dataset_path):
|
| 63 |
try:
|
| 64 |
-
return Image.open(dataset_path)
|
| 65 |
-
except:
|
| 66 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 67 |
|
| 68 |
-
# Create a placeholder image
|
| 69 |
placeholder = Image.new('RGB', (200, 200), color=(220, 220, 220))
|
| 70 |
from PIL import ImageDraw, ImageFont
|
| 71 |
draw = ImageDraw.Draw(placeholder)
|
| 72 |
# Try to use default font, fallback to basic if not available
|
| 73 |
try:
|
| 74 |
-
font = ImageFont.truetype("/usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf",
|
| 75 |
except:
|
| 76 |
font = ImageFont.load_default()
|
| 77 |
|
| 78 |
-
# Draw
|
| 79 |
-
|
| 80 |
-
|
| 81 |
-
|
| 82 |
-
|
| 83 |
-
|
| 84 |
-
|
|
|
|
|
|
|
| 85 |
|
| 86 |
-
|
|
|
|
| 87 |
|
| 88 |
# --- 3. Define the Search Logic ---
|
| 89 |
def find_best_matches(user_image):
|
|
@@ -121,18 +175,59 @@ def find_best_matches(user_image):
|
|
| 121 |
display_name = f"{row['name']} (Match: {int(score*100)}%)"
|
| 122 |
result_text += f"### #{i}: {display_name}\n\n"
|
| 123 |
|
| 124 |
-
# Load Image
|
| 125 |
try:
|
| 126 |
-
img =
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 127 |
gallery_images.append((img, display_name))
|
| 128 |
-
|
| 129 |
-
|
| 130 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 131 |
except Exception as img_error:
|
| 132 |
# Even if loading fails, create placeholder
|
| 133 |
placeholder = Image.new('RGB', (200, 200), color=(220, 220, 220))
|
| 134 |
gallery_images.append((placeholder, display_name))
|
| 135 |
-
result_text += f"β οΈ
|
| 136 |
|
| 137 |
# Don't pad with None - Gallery can't handle None images
|
| 138 |
# Just return what we have
|
|
|
|
| 27 |
if DB_PATH:
|
| 28 |
print(f"π Loaded Knowledge Base: {DB_PATH}")
|
| 29 |
df_db = pd.read_parquet(DB_PATH)
|
| 30 |
+
print(f"π Database columns: {df_db.columns.tolist()}")
|
| 31 |
+
print(f"π Database shape: {df_db.shape}")
|
| 32 |
# Convert embedding column to a clean numpy matrix for fast math
|
| 33 |
DB_VECTORS = np.stack(df_db['embedding'].values)
|
| 34 |
|
|
|
|
| 36 |
# e.g. "famous_faces_GhostFaceNet.parquet" -> "GhostFaceNet"
|
| 37 |
MODEL_NAME = DB_PATH.split("_")[-1].replace(".parquet", "")
|
| 38 |
print(f"βοΈ Model configured: {MODEL_NAME}")
|
| 39 |
+
|
| 40 |
+
# Debug: Check what directories exist
|
| 41 |
+
print(f"π Checking available directories...")
|
| 42 |
+
if os.path.exists("my_dataset"):
|
| 43 |
+
files_in_dataset = os.listdir("my_dataset")[:5] # Show first 5
|
| 44 |
+
print(f" Found 'my_dataset' directory with {len(os.listdir('my_dataset'))} files")
|
| 45 |
+
print(f" Sample files: {files_in_dataset}")
|
| 46 |
+
else:
|
| 47 |
+
print(f" 'my_dataset' directory not found")
|
| 48 |
+
|
| 49 |
+
# Debug: Show sample image paths from database
|
| 50 |
+
if 'image_path' in df_db.columns:
|
| 51 |
+
sample_paths = df_db['image_path'].head(3).tolist()
|
| 52 |
+
print(f"π· Sample image paths from DB: {sample_paths}")
|
| 53 |
else:
|
| 54 |
print("β CRITICAL: No Parquet file found! Please run the Model Battle step.")
|
| 55 |
DB_VECTORS = None
|
|
|
|
| 57 |
|
| 58 |
# --- 2. Helper function to load images with fallbacks ---
|
| 59 |
def load_image_with_fallback(image_path, name):
|
| 60 |
+
"""
|
| 61 |
+
Try to load image from various possible locations, or create placeholder.
|
| 62 |
+
|
| 63 |
+
The parquet file stores image_path (e.g., 'my_dataset/Sahar_Milis.png'),
|
| 64 |
+
but we need to find where the actual image files are located.
|
| 65 |
+
"""
|
| 66 |
+
tried_paths = []
|
| 67 |
+
|
| 68 |
+
# Strategy 1: Try original path as-is
|
| 69 |
if os.path.exists(image_path):
|
| 70 |
try:
|
| 71 |
+
return Image.open(image_path), image_path
|
| 72 |
+
except Exception as e:
|
| 73 |
+
tried_paths.append(f"{image_path} (error: {e})")
|
| 74 |
+
else:
|
| 75 |
+
tried_paths.append(f"{image_path} (not found)")
|
| 76 |
|
| 77 |
+
# Strategy 2: Try just the filename in current directory
|
| 78 |
filename = os.path.basename(image_path)
|
| 79 |
if os.path.exists(filename):
|
| 80 |
try:
|
| 81 |
+
return Image.open(filename), filename
|
| 82 |
+
except Exception as e:
|
| 83 |
+
tried_paths.append(f"{filename} (error: {e})")
|
| 84 |
+
else:
|
| 85 |
+
tried_paths.append(f"{filename} (not found)")
|
| 86 |
+
|
| 87 |
+
# Strategy 3: Try in my_dataset directory (remove my_dataset/ prefix if present)
|
| 88 |
+
if image_path.startswith("my_dataset/"):
|
| 89 |
+
# Already has prefix, try as-is
|
| 90 |
+
dataset_path = image_path
|
| 91 |
+
else:
|
| 92 |
+
# Add prefix
|
| 93 |
+
dataset_path = os.path.join("my_dataset", filename)
|
| 94 |
|
|
|
|
|
|
|
| 95 |
if os.path.exists(dataset_path):
|
| 96 |
try:
|
| 97 |
+
return Image.open(dataset_path), dataset_path
|
| 98 |
+
except Exception as e:
|
| 99 |
+
tried_paths.append(f"{dataset_path} (error: {e})")
|
| 100 |
+
else:
|
| 101 |
+
tried_paths.append(f"{dataset_path} (not found)")
|
| 102 |
+
|
| 103 |
+
# Strategy 4: Try searching in current directory recursively
|
| 104 |
+
import glob
|
| 105 |
+
search_patterns = [
|
| 106 |
+
f"**/{filename}",
|
| 107 |
+
f"**/*{filename}",
|
| 108 |
+
f"**/{name.replace(' ', '_')}*",
|
| 109 |
+
f"**/{name.replace(' ', '-')}*",
|
| 110 |
+
]
|
| 111 |
+
for pattern in search_patterns:
|
| 112 |
+
matches = glob.glob(pattern, recursive=True)
|
| 113 |
+
if matches:
|
| 114 |
+
try:
|
| 115 |
+
return Image.open(matches[0]), matches[0]
|
| 116 |
+
except:
|
| 117 |
+
continue
|
| 118 |
|
| 119 |
+
# Strategy 5: Create a placeholder image with name
|
| 120 |
placeholder = Image.new('RGB', (200, 200), color=(220, 220, 220))
|
| 121 |
from PIL import ImageDraw, ImageFont
|
| 122 |
draw = ImageDraw.Draw(placeholder)
|
| 123 |
# Try to use default font, fallback to basic if not available
|
| 124 |
try:
|
| 125 |
+
font = ImageFont.truetype("/usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf", 16)
|
| 126 |
except:
|
| 127 |
font = ImageFont.load_default()
|
| 128 |
|
| 129 |
+
# Draw name and "Not Found" text
|
| 130 |
+
text_lines = [name[:15], "Image", "Not Found"]
|
| 131 |
+
y_offset = 50
|
| 132 |
+
for line in text_lines:
|
| 133 |
+
bbox = draw.textbbox((0, 0), line, font=font)
|
| 134 |
+
text_width = bbox[2] - bbox[0]
|
| 135 |
+
position = ((200 - text_width) // 2, y_offset)
|
| 136 |
+
draw.text(position, line, fill=(100, 100, 100), font=font)
|
| 137 |
+
y_offset += 30
|
| 138 |
|
| 139 |
+
print(f"β οΈ Could not find image for {name}. Tried: {tried_paths[:3]}")
|
| 140 |
+
return placeholder, None
|
| 141 |
|
| 142 |
# --- 3. Define the Search Logic ---
|
| 143 |
def find_best_matches(user_image):
|
|
|
|
| 175 |
display_name = f"{row['name']} (Match: {int(score*100)}%)"
|
| 176 |
result_text += f"### #{i}: {display_name}\n\n"
|
| 177 |
|
| 178 |
+
# Load Image - prioritize bytes from parquet, then try file paths
|
| 179 |
try:
|
| 180 |
+
img = None
|
| 181 |
+
found_path = None
|
| 182 |
+
|
| 183 |
+
# Strategy 1: Check if image is stored as bytes in parquet (BEST - self-contained)
|
| 184 |
+
if 'image_bytes' in df_db.columns and row.get('image_bytes') is not None:
|
| 185 |
+
try:
|
| 186 |
+
import io
|
| 187 |
+
img = Image.open(io.BytesIO(row['image_bytes']))
|
| 188 |
+
found_path = "parquet (embedded)"
|
| 189 |
+
except Exception as e:
|
| 190 |
+
print(f"β οΈ Could not load image bytes for {row['name']}: {e}")
|
| 191 |
+
|
| 192 |
+
# Strategy 2: Try loading from file path (fallback)
|
| 193 |
+
if img is None and 'image_path' in df_db.columns:
|
| 194 |
+
img, found_path = load_image_with_fallback(row['image_path'], row['name'])
|
| 195 |
+
|
| 196 |
+
# Strategy 3: Create placeholder if still no image
|
| 197 |
+
if img is None:
|
| 198 |
+
placeholder = Image.new('RGB', (200, 200), color=(220, 220, 220))
|
| 199 |
+
from PIL import ImageDraw, ImageFont
|
| 200 |
+
draw = ImageDraw.Draw(placeholder)
|
| 201 |
+
try:
|
| 202 |
+
font = ImageFont.truetype("/usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf", 16)
|
| 203 |
+
except:
|
| 204 |
+
font = ImageFont.load_default()
|
| 205 |
+
text_lines = [row['name'][:15], "Image", "Not Found"]
|
| 206 |
+
y_offset = 50
|
| 207 |
+
for line in text_lines:
|
| 208 |
+
bbox = draw.textbbox((0, 0), line, font=font)
|
| 209 |
+
text_width = bbox[2] - bbox[0]
|
| 210 |
+
position = ((200 - text_width) // 2, y_offset)
|
| 211 |
+
draw.text(position, line, fill=(100, 100, 100), font=font)
|
| 212 |
+
y_offset += 30
|
| 213 |
+
img = placeholder
|
| 214 |
+
found_path = None
|
| 215 |
+
|
| 216 |
gallery_images.append((img, display_name))
|
| 217 |
+
|
| 218 |
+
# Add status message
|
| 219 |
+
if found_path == "parquet (embedded)":
|
| 220 |
+
result_text += f"β Image loaded from parquet\n\n"
|
| 221 |
+
elif found_path:
|
| 222 |
+
result_text += f"β Found image at: {found_path}\n\n"
|
| 223 |
+
else:
|
| 224 |
+
result_text += f"β οΈ Image not found for {row['name']}\n\n"
|
| 225 |
+
|
| 226 |
except Exception as img_error:
|
| 227 |
# Even if loading fails, create placeholder
|
| 228 |
placeholder = Image.new('RGB', (200, 200), color=(220, 220, 220))
|
| 229 |
gallery_images.append((placeholder, display_name))
|
| 230 |
+
result_text += f"β οΈ Error loading image: {str(img_error)}\n\n"
|
| 231 |
|
| 232 |
# Don't pad with None - Gallery can't handle None images
|
| 233 |
# Just return what we have
|
famous_faces_GhostFaceNet.parquet β famous_faces_ArcFace_standalone.parquet
RENAMED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:f5e212ec446c3014d737f327a5788deb262fcadd0bcee7c6052dda8a356727c4
|
| 3 |
+
size 12055699
|