laudari's picture
Update app.py
b601afa verified
import gradio as gr
from PIL import Image
import numpy as np
import cv2
import json
import albumentations as A
from typing import List, Tuple, Dict, Any
import supervision as sv
import uuid
import random
from pathlib import Path
import colorsys
import logging
import zipfile
import io
from datetime import datetime
# Set up logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)
class PolygonAugmentation:
def __init__(self, tolerance=0.2, area_threshold=0.01, debug=False):
self.tolerance = tolerance
self.area_threshold = area_threshold
self.debug = debug
self.supported_extensions = ['.png', '.jpg', '.jpeg', '.bmp', '.PNG', '.JPEG']
self.augmented_results = [] # Store all augmentation results
def __getattr__(self, name: str) -> Any:
raise AttributeError(f"'PolygonAugmentation' object has no attribute '{name}'")
def calculate_polygon_area(self, points: List[List[float]]) -> float:
poly_np = np.array(points, dtype=np.float32)
area = cv2.contourArea(poly_np)
if self.debug:
logger.info(f"[DEBUG] Calculating polygon area: {area:.2f}")
return area
def load_labelme_data(self, json_file: Any, image: np.ndarray) -> Tuple:
if isinstance(json_file, str):
with open(json_file, 'r', encoding='utf-8') as f:
data = json.load(f)
elif isinstance(json_file, dict):
# Handle dictionary data directly
data = json_file
else:
# Handle file object
data = json.load(json_file)
shapes = []
if 'shapes' in data and isinstance(data['shapes'], list):
shapes = data['shapes']
elif 'segments' in data and isinstance(data['segments'], list):
shapes = [
{
"label": seg.get("class", "unknown"),
"points": seg.get("polygon", []),
"shape_type": "polygon",
"group_id": None,
"flags": {},
"confidence": seg.get("confidence", 1.0)
}
for seg in data['segments']
]
else:
raise ValueError("Invalid JSON: Neither 'shapes' nor 'segments' key found or not a list")
polygons = []
labels = []
original_areas = []
for shape in shapes:
if shape.get('shape_type') != 'polygon' or not shape.get('points') or len(shape['points']) < 3:
if self.debug:
logger.info(f"[DEBUG] Skipping invalid shape: {shape}")
continue
try:
points = [[float(x), float(y)] for x, y in shape['points']]
polygons.append(points)
labels.append(shape['label'])
original_areas.append(self.calculate_polygon_area(points))
except (ValueError, TypeError) as e:
if self.debug:
logger.info(f"[DEBUG] Error processing points: {shape['points']}, error: {str(e)}")
continue
if not polygons and self.debug:
logger.info(f"[DEBUG] Warning: No valid polygons in JSON")
return image, polygons, labels, original_areas, data, "input"
def simplify_polygon(self, polygon: List[List[float]], tolerance: float = None, label: str = None) -> List[List[float]]:
tol = tolerance if tolerance is not None else self.tolerance
if label and label.lower() in ['background', 'bg', 'back']:
tol = tol * 3
if self.debug:
logger.info(f"[DEBUG] Using increased tolerance {tol} for background label '{label}'")
if len(polygon) < 3:
if self.debug:
logger.info(f"[DEBUG] Polygon has fewer than 3 points, skipping simplification.")
return polygon
poly_np = np.array(polygon, dtype=np.float32)
approx = cv2.approxPolyDP(poly_np, tol, closed=True)
simplified = approx.reshape(-1, 2).tolist()
if self.debug:
logger.info(f"[DEBUG] Simplified polygon from {len(polygon)} to {len(simplified)} points with tolerance {tol}")
return simplified
def create_donut_polygon(self, external_contour: np.ndarray, internal_contours: List[np.ndarray]) -> List[List[float]]:
"""Create a donut/ring polygon by connecting external and internal contours with bridges"""
external_points = external_contour.reshape(-1, 2).tolist()
if not internal_contours:
if self.debug:
logger.info("[DEBUG] No internal contours found, returning external points.")
return external_points
# Start with external contour points
result_points = external_points.copy()
# Process each internal contour (hole)
for hole_idx, internal_contour in enumerate(internal_contours):
internal_points = internal_contour.reshape(-1, 2).tolist()
# Find the closest point between external and internal contours
min_dist = float('inf')
best_ext_idx = 0
best_int_idx = 0
# Check all combinations to find minimum distance
for i, ext_point in enumerate(result_points):
for j, int_point in enumerate(internal_points):
dist = np.sqrt((ext_point[0] - int_point[0])**2 + (ext_point[1] - int_point[1])**2)
if dist < min_dist:
min_dist = dist
best_ext_idx = i
best_int_idx = j
# Create bridge points
bridge_start = result_points[best_ext_idx]
connect_point = internal_points[best_int_idx]
if self.debug:
logger.info(f"[DEBUG] Creating bridge for hole {hole_idx}: ext_idx={best_ext_idx}, int_idx={best_int_idx}, distance={min_dist:.2f}")
# Insert the internal contour into the result
# Order: external_points[:best_ext_idx+1] + internal_hole + back_to_external + external_points[best_ext_idx+1:]
new_result = (
result_points[:best_ext_idx+1] + # External points up to bridge
internal_points[best_int_idx:] + # Internal points from connection point to end
internal_points[:best_int_idx+1] + # Internal points from start to connection point
[bridge_start] + # Bridge back to external
result_points[best_ext_idx+1:] # Remaining external points
)
result_points = new_result
if self.debug:
logger.info(f"[DEBUG] Created donut polygon with {len(result_points)} total points")
return result_points
def save_augmented_data(
self,
aug_image: np.ndarray,
aug_polygons: List[List[List[float]]],
aug_labels: List[str],
original_data: Dict[str, Any],
base_name: str
) -> Dict[str, Any]:
aug_id = uuid.uuid4().hex[:4]
aug_img_name = f"{base_name}_{aug_id}_aug.png"
new_shapes = []
for poly, label in zip(aug_polygons, aug_labels):
if not poly or len(poly) < 3:
continue
# Create LabelMe format shape
shape_data = {
"label": label,
"points": poly,
"group_id": None,
"shape_type": "polygon",
"flags": {},
"description": "",
"attributes": {},
"iscrowd": 0,
"difficult": 0
}
# Add additional metadata for special polygon types
if label.lower() in ['ring', 'donut', 'annulus', 'circle', 'round']:
shape_data["attributes"]["polygon_type"] = "ring"
elif label.lower() in ['background', 'bg', 'back']:
shape_data["attributes"]["polygon_type"] = "background"
else:
shape_data["attributes"]["polygon_type"] = "object"
new_shapes.append(shape_data)
# Get actual dimensions from augmented image
aug_height, aug_width = aug_image.shape[:2]
# Create LabelMe compatible JSON structure
aug_data = {
"version": original_data.get("version", "5.0.1"),
"flags": original_data.get("flags", {}),
"shapes": new_shapes,
"imagePath": aug_img_name,
"imageData": None, # Explicitly set to None as requested
"imageHeight": aug_height,
"imageWidth": aug_width,
"imageDepth": 3 if len(aug_image.shape) == 3 else 1,
# Additional LabelMe metadata
"lineColor": [0, 255, 0, 128],
"fillColor": [255, 0, 0, 128],
"textSize": 10,
"textColor": [0, 0, 0, 255],
# Augmentation metadata
"augmentation": {
"augmented": True,
"augmentation_id": aug_id,
"original_file": original_data.get("imagePath", "unknown"),
"augmentation_timestamp": datetime.now().isoformat(),
"augmentation_tool": "PolygonAugmentation v1.0"
}
}
if self.debug:
logger.info(f"[DEBUG] Created LabelMe JSON: {len(new_shapes)} shapes, size: {aug_width}x{aug_height}")
logger.info(f"[DEBUG] Shape types: {[s['attributes'].get('polygon_type', 'unknown') for s in new_shapes]}")
return aug_data
def polygons_to_masks(self, image: np.ndarray, polygons: List[List[List[float]]], labels: List[str]) -> Tuple[np.ndarray, List[str]]:
height, width = image.shape[:2]
all_masks = []
all_labels = []
for poly_idx, (poly, label) in enumerate(zip(polygons, labels)):
try:
poly_np = np.array(poly, dtype=np.int32)
if len(poly_np) < 3:
if self.debug:
logger.info(f"[DEBUG] Skipping polygon {poly_idx}: fewer than 3 points")
continue
mask = np.zeros((height, width), dtype=np.uint8)
cv2.fillPoly(mask, [poly_np], 1)
all_masks.append(mask)
all_labels.append(label)
except Exception as e:
if self.debug:
logger.info(f"[DEBUG] Error processing polygon {poly_idx}: {str(e)}")
if not all_masks:
return np.zeros((0, height, width), dtype=np.uint8), []
return np.array(all_masks, dtype=np.uint8), all_labels
def process_contours(
self,
external_contour: np.ndarray,
internal_contours: List[np.ndarray],
width: int,
height: int,
label: str,
all_polygons: List[List[List[float]]],
all_labels: List[str],
tolerance: float = None
) -> None:
tol = tolerance if tolerance is not None else self.tolerance
external_points = external_contour.reshape(-1, 2).tolist()
simplified_external = self.simplify_polygon(external_points, tolerance=tol, label=label)
if len(simplified_external) >= 3:
poly_labelme = [[round(max(0, min(float(x), width - 1)), 2),
round(max(0, min(float(y), height - 1)), 2)]
for x, y in simplified_external]
all_polygons.append(poly_labelme)
all_labels.append(label)
if self.debug:
logger.info(f"[DEBUG] Added simplified external polygon with {len(poly_labelme)} points.")
for internal_contour in internal_contours:
internal_points = internal_contour.reshape(-1, 2).tolist()
simplified_internal = self.simplify_polygon(internal_points, tolerance=tol, label=label)
if len(simplified_internal) >= 3:
poly_labelme = [[round(max(0, min(float(x), width - 1)), 2),
round(max(0, min(float(y), height - 1)), 2)]
for x, y in simplified_internal]
all_polygons.append(poly_labelme)
all_labels.append(label)
if self.debug:
logger.info(f"[DEBUG] Added simplified internal polygon with {len(poly_labelme)} points.")
def masks_to_labelme_polygons(
self,
masks: np.ndarray,
labels: List[str],
original_areas: List[float],
area_threshold: float = None,
tolerance: float = None
) -> Tuple[List[List[List[float]]], List[str]]:
tol = tolerance if tolerance is not None else self.tolerance
area_thresh = area_threshold if area_threshold is not None else self.area_threshold
height, width = masks[0].shape if len(masks) > 0 else (0, 0)
all_polygons = []
all_labels = []
for mask_idx, (mask, label) in enumerate(zip(masks, labels)):
if mask.sum() < 10:
if self.debug:
logger.info(f"[DEBUG] Skipping mask {mask_idx}: very small or empty.")
continue
contours, hierarchy = cv2.findContours(mask, cv2.RETR_CCOMP, cv2.CHAIN_APPROX_SIMPLE)
if hierarchy is None or len(contours) == 0:
if self.debug:
logger.info(f"[DEBUG] No contours found in mask {mask_idx}.")
continue
hierarchy = hierarchy[0]
external_contours = []
internal_contours_map = {}
for i, (contour, h) in enumerate(zip(contours, hierarchy)):
if h[3] == -1:
external_contours.append(contour)
internal_contours_map[len(external_contours)-1] = []
else:
parent_idx = h[3]
for j, _ in enumerate(external_contours):
if parent_idx == j:
internal_contours_map[j].append(contour)
break
if not external_contours:
if self.debug:
logger.info(f"[DEBUG] No external contours found in mask {mask_idx}.")
continue
for ext_idx, external_contour in enumerate(external_contours):
internal_contours = internal_contours_map.get(ext_idx, [])
ext_area = cv2.contourArea(external_contour)
if ext_area <= 0:
continue
if mask_idx < len(original_areas) and original_areas[mask_idx] > 0:
relative_area = ext_area / original_areas[mask_idx]
if relative_area < area_thresh:
if self.debug:
logger.info(f"[DEBUG] Skipping contour {ext_idx} (area too small: {relative_area:.4f})")
continue
# Check if this is a ring/donut shape or complex polygon
is_ring_shape = label.lower() in ['ring', 'donut', 'annulus', 'circle', 'round'] or len(internal_contours) > 0
is_background = label.lower() in ['background', 'bg', 'back']
# Handle different polygon types
if (is_background or is_ring_shape) and internal_contours:
try:
# Create donut polygon for rings, backgrounds, or shapes with holes
donut_points = self.create_donut_polygon(external_contour, internal_contours)
simplified_donut = self.simplify_polygon(donut_points, tolerance=tol, label=label)
if len(simplified_donut) >= 3:
# Ensure all points are within image boundaries
poly_labelme = []
for x, y in simplified_donut:
clipped_x = round(max(0, min(float(x), width - 1)), 2)
clipped_y = round(max(0, min(float(y), height - 1)), 2)
poly_labelme.append([clipped_x, clipped_y])
all_polygons.append(poly_labelme)
all_labels.append(label)
if self.debug:
logger.info(f"[DEBUG] Added {'ring' if is_ring_shape else 'background'} donut polygon with {len(poly_labelme)} points, {len(internal_contours)} holes")
else:
if self.debug:
logger.info(f"[DEBUG] Donut polygon too small after simplification, falling back to separate contours")
# Fallback to separate contours
self.process_contours(
external_contour, internal_contours, width, height,
label, all_polygons, all_labels, tol
)
except Exception as e:
if self.debug:
logger.info(f"[DEBUG] Error creating donut for {label}: {str(e)}, fallback to separate polygons.")
# Fallback to processing contours separately
self.process_contours(
external_contour, internal_contours, width, height,
label, all_polygons, all_labels, tol
)
else:
# Handle regular polygons (no holes or simple shapes)
self.process_contours(
external_contour, internal_contours, width, height,
label, all_polygons, all_labels, tol
)
return all_polygons, all_labels
def augment_single_image(
self,
image: np.ndarray,
polygons: List[List[List[float]]],
labels: List[str],
original_areas: List[float],
original_data: Dict[str, Any],
aug_type: str,
aug_param: float
) -> Tuple[np.ndarray, Dict[str, Any]]:
logger.info(f"Applying augmentation: {aug_type} with parameter {aug_param}")
height, width = image.shape[:2]
# Setup augmentation based on type with proper parameters
if aug_type == "rotate":
# For rotation, use the parameter as degrees and make it more visible
rotation_angle = aug_param if abs(aug_param) >= 5 else (15 if aug_param >= 0 else -15)
# Use angle directly (not abs) and set limit as tuple for specific angle
aug_transform = A.Rotate(limit=(rotation_angle, rotation_angle), p=1.0, border_mode=cv2.BORDER_CONSTANT, value=0)
logger.info(f"Applying rotation: {rotation_angle} degrees")
elif aug_type == "horizontal_flip":
aug_transform = A.HorizontalFlip(p=1.0 if aug_param == 1 else 0.0)
elif aug_type == "vertical_flip":
aug_transform = A.VerticalFlip(p=1.0 if aug_param == 1 else 0.0)
elif aug_type == "scale":
# Ensure scale parameter is reasonable
scale_factor = max(0.5, min(2.0, aug_param))
aug_transform = A.Affine(scale=scale_factor, p=1.0, keep_ratio=True)
logger.info(f"Applying scale: {scale_factor}")
elif aug_type == "brightness_contrast":
brightness_factor = max(-0.5, min(0.5, aug_param))
aug_transform = A.RandomBrightnessContrast(
brightness_limit=abs(brightness_factor),
contrast_limit=abs(brightness_factor),
p=1.0
)
elif aug_type == "pixel_dropout":
dropout_prob = min(max(aug_param, 0.0), 0.2)
aug_transform = A.PixelDropout(dropout_prob=dropout_prob, p=1.0)
else:
raise ValueError(f"Unsupported augmentation type: {aug_type}")
# Create masks from polygons
masks, mask_labels = self.polygons_to_masks(image, polygons, labels)
if masks.shape[0] == 0:
raise ValueError("No valid masks created from polygons")
# Convert masks array to list for albumentations
masks_list = [masks[i] for i in range(masks.shape[0])]
# Create additional targets for each mask
additional_targets = {f'mask{i}': 'mask' for i in range(len(masks_list))}
# Create transform with proper mask handling
transform = A.Compose([
aug_transform
], additional_targets=additional_targets)
# Prepare input dictionary
input_dict = {'image': image}
for i, mask in enumerate(masks_list):
input_dict[f'mask{i}'] = mask
# Apply augmentation
aug_result = transform(**input_dict)
aug_image = aug_result['image']
# Collect augmented masks and ensure they match image dimensions
aug_masks_list = []
aug_height, aug_width = aug_image.shape[:2]
for i in range(len(masks_list)):
aug_mask = aug_result[f'mask{i}']
# Ensure mask dimensions match augmented image
if aug_mask.shape[:2] != (aug_height, aug_width):
aug_mask = cv2.resize(aug_mask, (aug_width, aug_height), interpolation=cv2.INTER_NEAREST)
aug_masks_list.append(aug_mask)
aug_masks = np.array(aug_masks_list, dtype=np.uint8)
# Validate augmented image
if aug_image is None or aug_image.size == 0:
raise ValueError("Augmented image is empty or invalid")
# Convert augmented masks back to polygons
aug_polygons, aug_labels = self.masks_to_labelme_polygons(
aug_masks, mask_labels, original_areas, self.area_threshold, self.tolerance
)
# Apply random crop as post-processing to add variety
if random.random() < 0.3: # 30% chance of cropping
crop_scale = random.uniform(0.85, 0.95)
crop_height = int(aug_height * crop_scale)
crop_width = int(aug_width * crop_scale)
# Create crop transform
crop_transform = A.Compose([
A.RandomCrop(width=crop_width, height=crop_height, p=1.0)
], additional_targets={f'mask{i}': 'mask' for i in range(len(aug_masks_list))})
# Apply crop
crop_input = {'image': aug_image}
for i, mask in enumerate(aug_masks_list):
crop_input[f'mask{i}'] = mask
crop_result = crop_transform(**crop_input)
aug_image = crop_result['image']
# Update masks after crop
cropped_masks = []
for i in range(len(aug_masks_list)):
cropped_masks.append(crop_result[f'mask{i}'])
aug_masks = np.array(cropped_masks, dtype=np.uint8)
# Re-convert masks to polygons after crop
aug_polygons, aug_labels = self.masks_to_labelme_polygons(
aug_masks, mask_labels, original_areas, self.area_threshold, self.tolerance
)
# Create augmented data with correct dimensions
aug_data = self.save_augmented_data(aug_image, aug_polygons, aug_labels, original_data, "input")
logger.info(f"Augmentation completed: {len(aug_polygons)} polygons generated, final size: {aug_image.shape[:2]}")
return aug_image, aug_data
def batch_augment_images(self, image_json_pairs, aug_configs, num_augmentations):
"""Batch process multiple images with multiple augmentation configurations"""
logger.info(f"Starting batch augmentation with {len(image_json_pairs)} pairs, {len(aug_configs)} configs, {num_augmentations} augmentations each")
self.augmented_results = []
results = []
for pair_idx, (image, json_data) in enumerate(image_json_pairs):
if image is None or json_data is None:
logger.warning(f"Skipping pair {pair_idx}: missing image or JSON data")
continue
try:
logger.info(f"Processing image pair {pair_idx}")
# Convert PIL image to NumPy
img_np = np.array(image)
img_np = cv2.cvtColor(img_np, cv2.COLOR_RGB2BGR)
# Load data - pass the JSON data directly
img_np, polygons, labels, original_areas, original_data, _ = self.load_labelme_data(json_data, img_np)
logger.info(f"Loaded {len(polygons)} polygons for image {pair_idx}")
# Apply each augmentation configuration
for config_idx, config in enumerate(aug_configs):
logger.info(f"Applying config {config_idx}: {config['aug_type']}")
for aug_idx in range(num_augmentations):
# Generate random parameter within range
min_val, max_val = config['param_range']
if config['aug_type'] in ['horizontal_flip', 'vertical_flip']:
aug_param = random.choice([0, 1])
else:
aug_param = random.uniform(min_val, max_val)
try:
logger.info(f"Generating augmentation {aug_idx} with {config['aug_type']}, param: {aug_param}")
aug_image, aug_data = self.augment_single_image(
img_np, polygons, labels, original_areas,
original_data, config['aug_type'], aug_param
)
# Create visualization
aug_image_vis = self.create_visualization(aug_image, aug_data)
# Store result
result_data = {
'image': aug_image_vis,
'json_data': aug_data,
'metadata': {
'original_image_index': pair_idx,
'augmentation_index': aug_idx,
'augmentation_type': config['aug_type'],
'parameter_value': aug_param,
'parameter_range': config['param_range'],
'timestamp': datetime.now().isoformat(),
'filename': f'aug_{pair_idx}_{config["aug_type"]}_{aug_idx}.png'
}
}
self.augmented_results.append(result_data)
results.append(aug_image_vis)
logger.info(f"Successfully generated augmentation {aug_idx} for image {pair_idx}")
except Exception as e:
logger.error(f"Error augmenting image {pair_idx} with {config['aug_type']}: {str(e)}")
import traceback
logger.error(traceback.format_exc())
continue
except Exception as e:
logger.error(f"Error processing image pair {pair_idx}: {str(e)}")
import traceback
logger.error(traceback.format_exc())
continue
logger.info(f"Batch augmentation completed. Generated {len(results)} total results.")
return results
def create_visualization(self, aug_image, aug_data):
"""Create visualization with colored polygon masks and outlines for each class"""
# Create a dynamic color map for unique labels with better color distribution
unique_labels = list(set(shape['label'] for shape in aug_data['shapes']))
if not unique_labels:
label_color_map = {"unknown": (0, 255, 0)}
else:
num_labels = len(unique_labels)
# Create more distinct colors using different hue ranges
label_color_map = {}
for i, label in enumerate(unique_labels):
if label.lower() in ['background', 'bg', 'back']:
# Background gets a neutral gray-blue color
rgb = (100, 149, 237) # Cornflower blue with low opacity
elif 'ring' in label.lower() or 'donut' in label.lower():
# Ring/donut shapes get purple-pink colors
hue = 0.8 + (i * 0.1) % 0.2 # Purple range
rgb = colorsys.hsv_to_rgb(hue, 0.8, 0.9)
rgb = tuple(int(c * 255) for c in rgb)
else:
# Regular objects get distributed colors across the spectrum
hue = (i * 0.618033988749895) % 1.0 # Golden ratio for better distribution
saturation = 0.7 + (i % 3) * 0.1 # Vary saturation
value = 0.8 + (i % 2) * 0.15 # Vary brightness
rgb = colorsys.hsv_to_rgb(hue, saturation, value)
rgb = tuple(int(c * 255) for c in rgb)
label_color_map[label] = rgb
# Convert augmented image to RGB for visualization
aug_image_rgb = cv2.cvtColor(aug_image, cv2.COLOR_BGR2RGB)
overlay = aug_image_rgb.copy()
height, width = aug_image.shape[:2]
# Create a composite mask to handle overlapping polygons
composite_mask = np.zeros((height, width, 3), dtype=np.uint8)
# Group shapes by label for better visualization
shapes_by_label = {}
for shape in aug_data['shapes']:
label = shape['label']
if label not in shapes_by_label:
shapes_by_label[label] = []
shapes_by_label[label].append(shape)
# Process each label group
for label, shapes in shapes_by_label.items():
color = label_color_map.get(label, (0, 255, 0))
# Create mask for all polygons of this label
label_mask = np.zeros((height, width), dtype=np.uint8)
for shape in shapes:
points = np.array(shape['points'], dtype=np.int32)
if len(points) < 3:
continue
# Fill the polygon area
cv2.fillPoly(label_mask, [points], 255)
# Apply color to the mask areas
if label_mask.sum() > 0: # Only if mask has content
# Determine alpha based on label type
if label.lower() in ['background', 'bg', 'back']:
alpha = 0.15 # Lower opacity for background
elif 'ring' in label.lower() or 'donut' in label.lower():
alpha = 0.4 # Medium opacity for rings
else:
alpha = 0.35 # Standard opacity for objects
# Create colored mask
colored_mask = np.zeros_like(aug_image_rgb)
colored_mask[label_mask == 255] = color
# Blend with overlay
mask_area = label_mask == 255
overlay[mask_area] = cv2.addWeighted(
overlay[mask_area],
1.0 - alpha,
colored_mask[mask_area],
alpha,
0
)
# Draw polygon outlines with thicker lines for better visibility
for shape in aug_data['shapes']:
label = shape['label']
color = label_color_map.get(label, (0, 255, 0))
points = np.array(shape['points'], dtype=np.int32)
if len(points) < 3:
continue
# Determine line thickness based on polygon type
if label.lower() in ['background', 'bg', 'back']:
thickness = 1 # Thinner lines for background
elif 'ring' in label.lower() or 'donut' in label.lower():
thickness = 3 # Thicker lines for rings to show structure
else:
thickness = 2 # Standard thickness
# Draw polygon outline
cv2.polylines(overlay, [points], isClosed=True, color=color, thickness=thickness)
# Add label text near the polygon
if len(points) > 0:
# Find a good position for the label
moments = cv2.moments(points)
if moments['m00'] != 0:
cx = int(moments['m10'] / moments['m00'])
cy = int(moments['m01'] / moments['m00'])
else:
cx, cy = points[0][0], points[0][1]
# Ensure text position is within image bounds
cx = max(10, min(cx, width - 50))
cy = max(20, min(cy, height - 10))
# Add text background for better readability
font = cv2.FONT_HERSHEY_SIMPLEX
font_scale = 0.4
text_thickness = 1
text_size = cv2.getTextSize(label, font, font_scale, text_thickness)[0]
# Draw background rectangle
cv2.rectangle(overlay,
(cx - 2, cy - text_size[1] - 4),
(cx + text_size[0] + 2, cy + 2),
(0, 0, 0), -1)
# Draw text
cv2.putText(overlay, label, (cx, cy - 2), font, font_scale, color, text_thickness)
if self.debug:
logger.info(f"[DEBUG] Created visualization with {len(unique_labels)} unique labels: {list(unique_labels)}")
return Image.fromarray(overlay)
def create_download_package(self):
"""Create a zip file with all augmented images and proper LabelMe JSON files"""
if not self.augmented_results:
logger.warning("No augmented results available for download")
return None
logger.info(f"Creating download package with {len(self.augmented_results)} results")
zip_buffer = io.BytesIO()
try:
with zipfile.ZipFile(zip_buffer, 'w', zipfile.ZIP_DEFLATED) as zip_file:
# Add all augmented images and their corresponding LabelMe JSON files
for idx, result in enumerate(self.augmented_results):
filename = result['metadata']['filename']
# Save augmented image
try:
# Convert PIL image to RGB if needed
if result['image'].mode != 'RGB':
img_rgb = result['image'].convert('RGB')
else:
img_rgb = result['image']
# Save as PNG bytes
img_buffer = io.BytesIO()
img_rgb.save(img_buffer, format='PNG', optimize=True)
zip_file.writestr(filename, img_buffer.getvalue())
logger.info(f"Added image: {filename}")
except Exception as e:
logger.error(f"Error saving image {filename}: {str(e)}")
continue
# Save corresponding LabelMe JSON file
json_filename = filename.replace('.png', '.json')
try:
# Create a clean LabelMe JSON structure
clean_json_data = {
"version": "5.0.1",
"flags": {},
"shapes": [],
"imagePath": filename,
"imageData": None, # No embedded image data as requested
"imageHeight": result['json_data']['imageHeight'],
"imageWidth": result['json_data']['imageWidth'],
"imageDepth": 3
}
# Copy shapes with proper LabelMe format
for shape in result['json_data']['shapes']:
clean_shape = {
"label": shape['label'],
"points": shape['points'],
"group_id": shape.get('group_id'),
"shape_type": "polygon",
"flags": shape.get('flags', {}),
"description": shape.get('description', ''),
"iscrowd": shape.get('iscrowd', 0),
"attributes": shape.get('attributes', {})
}
clean_json_data['shapes'].append(clean_shape)
# Write JSON file
json_str = json.dumps(clean_json_data, indent=2, ensure_ascii=False)
zip_file.writestr(json_filename, json_str)
logger.info(f"Added JSON: {json_filename} with {len(clean_json_data['shapes'])} shapes")
except Exception as e:
logger.error(f"Error saving JSON {json_filename}: {str(e)}")
continue
# Add comprehensive summary metadata
summary = {
'package_info': {
'total_augmentations': len(self.augmented_results),
'generation_timestamp': datetime.now().isoformat(),
'generator': 'PolygonAugmentation v1.0',
'format': 'LabelMe JSON + PNG images'
},
'augmentation_summary': [
{
'filename': result['metadata']['filename'],
'json_file': result['metadata']['filename'].replace('.png', '.json'),
'augmentation_type': result['metadata']['augmentation_type'],
'parameter_value': result['metadata']['parameter_value'],
'polygon_count': len(result['json_data']['shapes']),
'image_size': f"{result['json_data']['imageWidth']}x{result['json_data']['imageHeight']}",
'timestamp': result['metadata']['timestamp'],
'labels': list(set([shape['label'] for shape in result['json_data']['shapes']]))
}
for result in self.augmented_results
],
'statistics': {
'unique_augmentation_types': list(set([r['metadata']['augmentation_type'] for r in self.augmented_results])),
'total_polygons': sum([len(r['json_data']['shapes']) for r in self.augmented_results]),
'unique_labels': list(set([
shape['label']
for result in self.augmented_results
for shape in result['json_data']['shapes']
])),
'average_polygons_per_image': sum([len(r['json_data']['shapes']) for r in self.augmented_results]) / len(self.augmented_results) if self.augmented_results else 0
}
}
zip_file.writestr('augmentation_summary.json', json.dumps(summary, indent=2, ensure_ascii=False))
# Add README for the package
readme_content = f"""# Augmented Dataset Package
## Overview
This package contains {len(self.augmented_results)} augmented images with their corresponding LabelMe annotation files.
## Contents
- **Images**: PNG format augmented images
- **Annotations**: LabelMe JSON format annotation files (standard format)
- **Summary**: augmentation_summary.json with detailed metadata
## File Structure
- Each image file (*.png) has a corresponding annotation file (*.json) with the same base name
- All annotations are in standard LabelMe format without embedded image data
- Compatible with LabelMe, CVAT, and other annotation tools
## Statistics
- Total augmented images: {len(self.augmented_results)}
- Total polygons: {sum([len(r['json_data']['shapes']) for r in self.augmented_results])}
- Unique labels: {list(set([shape['label'] for result in self.augmented_results for shape in result['json_data']['shapes']]))}
- Augmentation types used: {list(set([r['metadata']['augmentation_type'] for r in self.augmented_results]))}
## Usage
1. Extract the ZIP file
2. Load images and annotations using any tool that supports LabelMe format
3. Use the augmentation_summary.json for batch processing or analysis
Generated on: {datetime.now().strftime('%Y-%m-%d %H:%M:%S')}
Tool: PolygonAugmentation v1.0
"""
zip_file.writestr('README.md', readme_content)
logger.info("Successfully created ZIP package with all files")
zip_buffer.seek(0)
logger.info(f"Created download package with {len(self.augmented_results)} image-annotation pairs")
return zip_buffer.getvalue()
except Exception as e:
logger.error(f"Error creating ZIP package: {str(e)}")
import traceback
logger.error(traceback.format_exc())
return None
def create_interface():
augmenter = PolygonAugmentation(tolerance=2.0, area_threshold=0.01, debug=True)
def process_batch_augmentation(
images, json_files, num_augmentations,
rotate_enabled, rotate_min, rotate_max,
hflip_enabled, vflip_enabled,
scale_enabled, scale_min, scale_max,
brightness_enabled, brightness_min, brightness_max,
dropout_enabled, dropout_min, dropout_max
):
if not images or not json_files:
return [], "No images or JSON files uploaded", None
# Pair images with JSON files
image_json_pairs = []
min_length = min(len(images), len(json_files))
for i in range(min_length):
if images[i] is not None and json_files[i] is not None:
try:
image = Image.open(images[i].name)
# Load JSON file content properly
json_path = json_files[i].name
logger.info(f"Loading JSON from: {json_path}")
with open(json_path, 'r', encoding='utf-8') as f:
json_data = json.load(f)
logger.info(f"Successfully loaded JSON with keys: {list(json_data.keys())}")
image_json_pairs.append((image, json_data))
except Exception as e:
logger.error(f"Error loading image/JSON pair {i}: {str(e)}")
import traceback
logger.error(traceback.format_exc())
continue
if not image_json_pairs:
return [], "No valid image-JSON pairs found", None
# Configure augmentations based on user selections
aug_configs = []
if rotate_enabled:
aug_configs.append({
'aug_type': 'rotate',
'param_range': (rotate_min, rotate_max)
})
if hflip_enabled:
aug_configs.append({
'aug_type': 'horizontal_flip',
'param_range': (0, 1)
})
if vflip_enabled:
aug_configs.append({
'aug_type': 'vertical_flip',
'param_range': (0, 1)
})
if scale_enabled:
aug_configs.append({
'aug_type': 'scale',
'param_range': (scale_min, scale_max)
})
if brightness_enabled:
aug_configs.append({
'aug_type': 'brightness_contrast',
'param_range': (brightness_min, brightness_max)
})
if dropout_enabled:
aug_configs.append({
'aug_type': 'pixel_dropout',
'param_range': (dropout_min, dropout_max)
})
if not aug_configs:
return [], "No augmentation types selected", None
# Process augmentations
try:
logger.info(f"Starting batch augmentation with {len(image_json_pairs)} image pairs and {len(aug_configs)} configurations")
augmented_images = augmenter.batch_augment_images(
image_json_pairs, aug_configs, num_augmentations
)
# Create JSON summary
json_summary = json.dumps([result['metadata'] for result in augmenter.augmented_results], indent=2)
status = f"Generated {len(augmented_images)} augmented images from {len(image_json_pairs)} input pairs"
logger.info(status)
return augmented_images, json_summary, status
except Exception as e:
error_msg = f"Batch augmentation error: {str(e)}"
logger.error(error_msg)
import traceback
logger.error(traceback.format_exc())
return [], error_msg, None
def download_package():
"""Handle download package creation and return proper file data"""
try:
package_data = augmenter.create_download_package()
if package_data is None:
return None
# Save the package to a temporary file for download
import tempfile
import os
# Create temporary file with proper name
temp_file = tempfile.NamedTemporaryFile(
delete=False,
suffix='.zip',
prefix='augmented_dataset_'
)
with open(temp_file.name, 'wb') as f:
f.write(package_data)
logger.info(f"Created download package: {temp_file.name}")
return temp_file.name
except Exception as e:
logger.error(f"Error creating download package: {str(e)}")
import traceback
logger.error(traceback.format_exc())
return None
def show_mask_overlay(evt: gr.SelectData):
if evt.index < len(augmenter.augmented_results):
return augmenter.augmented_results[evt.index]['image']
return None
with gr.Blocks(title="Dynamic Donut Polygon Augmentation") as demo:
gr.Markdown("# πŸŒ€ Dynamic Donut Polygon Augmentation Tool")
gr.Markdown("Upload multiple images and JSON files to apply batch augmentation with configurable parameter ranges")
with gr.Row():
with gr.Column(scale=1):
gr.Markdown("## πŸ“ Input Files")
images_input = gr.File(
file_count="multiple",
file_types=["image"],
label="Upload Images"
)
json_input = gr.File(
file_count="multiple",
file_types=[".json"],
label="Upload LabelMe JSON Files"
)
num_augmentations = gr.Slider(
minimum=1, maximum=5, value=2, step=1,
label="Augmentations per configuration"
)
gr.Markdown("## βš™οΈ Augmentation Configuration")
# Rotation parameters
with gr.Group():
rotate_enabled = gr.Checkbox(label="Enable Rotation", value=True)
with gr.Row():
rotate_min = gr.Slider(-45, 45, -15, label="Min Rotation (degrees)")
rotate_max = gr.Slider(-45, 45, 15, label="Max Rotation (degrees)")
# Flip parameters
with gr.Group():
hflip_enabled = gr.Checkbox(label="Enable Horizontal Flip", value=True)
vflip_enabled = gr.Checkbox(label="Enable Vertical Flip", value=False)
# Scale parameters
with gr.Group():
scale_enabled = gr.Checkbox(label="Enable Scale", value=True)
with gr.Row():
scale_min = gr.Slider(0.7, 1.3, 0.9, label="Min Scale")
scale_max = gr.Slider(0.7, 1.3, 1.1, label="Max Scale")
# Brightness parameters
with gr.Group():
brightness_enabled = gr.Checkbox(label="Enable Brightness/Contrast", value=True)
with gr.Row():
brightness_min = gr.Slider(-0.3, 0.3, -0.1, label="Min Brightness")
brightness_max = gr.Slider(-0.3, 0.3, 0.1, label="Max Brightness")
# Dropout parameters
with gr.Group():
dropout_enabled = gr.Checkbox(label="Enable Pixel Dropout", value=False)
with gr.Row():
dropout_min = gr.Slider(0.01, 0.1, 0.02, label="Min Dropout")
dropout_max = gr.Slider(0.01, 0.1, 0.05, label="Max Dropout")
generate_btn = gr.Button("πŸš€ Generate Augmentations", variant="primary")
status_text = gr.Textbox(label="Status", interactive=False)
with gr.Column(scale=2):
gr.Markdown("## πŸ–ΌοΈ Augmented Results")
gr.Markdown("*Click on any image to view with enhanced mask overlay*")
augmented_gallery = gr.Gallery(
label="Augmented Images with Polygon Masks",
show_label=False,
elem_id="gallery",
columns=3,
rows=3,
height="auto"
)
with gr.Row():
download_btn = gr.Button("πŸ“₯ Download All (ZIP)", variant="secondary")
download_file = gr.File(label="Download Package", visible=True)
gr.Markdown("## πŸ“‹ Augmentation Metadata")
json_output = gr.Code(
label="Generated Metadata JSON",
language="json",
lines=15
)
gr.Markdown("## 🎭 Enhanced Preview")
mask_preview = gr.Image(label="Selected Image with Mask Overlay")
# Event handlers
generate_btn.click(
process_batch_augmentation,
inputs=[
images_input, json_input, num_augmentations,
rotate_enabled, rotate_min, rotate_max,
hflip_enabled, vflip_enabled,
scale_enabled, scale_min, scale_max,
brightness_enabled, brightness_min, brightness_max,
dropout_enabled, dropout_min, dropout_max
],
outputs=[augmented_gallery, json_output, status_text]
)
download_btn.click(
download_package,
outputs=download_file
)
augmented_gallery.select(
show_mask_overlay,
outputs=mask_preview
)
return demo
if __name__ == "__main__":
demo = create_interface()
demo.launch()