Datasets:
Create parsing_code.py
Browse files- parsing_code.py +233 -0
parsing_code.py
ADDED
@@ -0,0 +1,233 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import json
|
3 |
+
import fiftyone as fo
|
4 |
+
from PIL import Image
|
5 |
+
from pathlib import Path
|
6 |
+
|
7 |
+
def load_sample_files(subdir):
|
8 |
+
"""
|
9 |
+
Load all required files for a single sample.
|
10 |
+
|
11 |
+
Args:
|
12 |
+
subdir (Path): Path to the sample subdirectory
|
13 |
+
|
14 |
+
Returns:
|
15 |
+
tuple: (detections_data, questions_data, mask_file_path, source_file_path, img_dimensions)
|
16 |
+
Returns None if any required files are missing.
|
17 |
+
"""
|
18 |
+
subdir_name = subdir.name
|
19 |
+
|
20 |
+
# Define file paths
|
21 |
+
detection_file = subdir / "detection.json"
|
22 |
+
question_file = subdir / "question.json"
|
23 |
+
mask_file = subdir / f"mask_{subdir_name}.png"
|
24 |
+
source_file = subdir / f"source_{subdir_name}.jpg"
|
25 |
+
|
26 |
+
# Check all files exist
|
27 |
+
if not all(f.exists() for f in [detection_file, question_file, mask_file, source_file]):
|
28 |
+
return None
|
29 |
+
|
30 |
+
# Load JSON data
|
31 |
+
with open(detection_file, 'r') as f:
|
32 |
+
detections_data = json.load(f)
|
33 |
+
|
34 |
+
with open(question_file, 'r') as f:
|
35 |
+
questions_data = json.load(f)
|
36 |
+
|
37 |
+
# Get image dimensions
|
38 |
+
with Image.open(source_file) as img:
|
39 |
+
img_dimensions = img.size
|
40 |
+
|
41 |
+
return detections_data, questions_data, mask_file, source_file, img_dimensions
|
42 |
+
|
43 |
+
def convert_detections_to_relative(detections_data, img_width, img_height):
|
44 |
+
"""
|
45 |
+
Convert absolute bounding boxes to relative coordinates for FiftyOne.
|
46 |
+
|
47 |
+
Args:
|
48 |
+
detections_data (list): List of detection dictionaries
|
49 |
+
img_width (int): Image width in pixels
|
50 |
+
img_height (int): Image height in pixels
|
51 |
+
|
52 |
+
Returns:
|
53 |
+
fo.Detections: FiftyOne Detections object
|
54 |
+
"""
|
55 |
+
detections = []
|
56 |
+
|
57 |
+
for detection_dict in detections_data:
|
58 |
+
for label, bbox in detection_dict.items():
|
59 |
+
x, y, width, height = bbox
|
60 |
+
# Convert to relative coordinates
|
61 |
+
rel_x = x / img_width
|
62 |
+
rel_y = y / img_height
|
63 |
+
rel_width = width / img_width
|
64 |
+
rel_height = height / img_height
|
65 |
+
|
66 |
+
detection = fo.Detection(
|
67 |
+
label=label,
|
68 |
+
bounding_box=[rel_x, rel_y, rel_width, rel_height]
|
69 |
+
)
|
70 |
+
detections.append(detection)
|
71 |
+
|
72 |
+
return fo.Detections(detections=detections)
|
73 |
+
|
74 |
+
def add_sample_metadata(sample, english_questions):
|
75 |
+
"""
|
76 |
+
Add sample-level metadata from questions data.
|
77 |
+
|
78 |
+
Args:
|
79 |
+
sample (fo.Sample): FiftyOne sample to modify
|
80 |
+
english_questions (list): List of English question dictionaries
|
81 |
+
"""
|
82 |
+
if not english_questions:
|
83 |
+
return
|
84 |
+
|
85 |
+
# Sample-level metadata (same for all questions in a sample)
|
86 |
+
first_question = english_questions[0]
|
87 |
+
sample['location'] = fo.Classification(label=first_question['location'])
|
88 |
+
sample['modality'] = fo.Classification(label=first_question['modality'])
|
89 |
+
sample['base_type'] = fo.Classification(label=first_question['base_type'])
|
90 |
+
sample['answer_type'] = fo.Classification(label=first_question['answer_type'])
|
91 |
+
|
92 |
+
def add_questions_and_answers(sample, english_questions):
|
93 |
+
"""
|
94 |
+
Add individual questions and answers to the sample.
|
95 |
+
|
96 |
+
Args:
|
97 |
+
sample (fo.Sample): FiftyOne sample to modify
|
98 |
+
english_questions (list): List of English question dictionaries
|
99 |
+
"""
|
100 |
+
for i, q_data in enumerate(english_questions):
|
101 |
+
sample[f'question_{i}'] = q_data['question']
|
102 |
+
sample[f'answer_{i}'] = fo.Classification(label=q_data['answer'])
|
103 |
+
|
104 |
+
def process_single_sample(subdir):
|
105 |
+
"""
|
106 |
+
Process a single sample directory into a FiftyOne sample.
|
107 |
+
|
108 |
+
Args:
|
109 |
+
subdir (Path): Path to the sample subdirectory
|
110 |
+
|
111 |
+
Returns:
|
112 |
+
fo.Sample or None: FiftyOne sample, or None if processing failed
|
113 |
+
"""
|
114 |
+
subdir_name = subdir.name
|
115 |
+
|
116 |
+
# Load all files for this sample
|
117 |
+
file_data = load_sample_files(subdir)
|
118 |
+
if file_data is None:
|
119 |
+
print(f"Warning: Missing files in {subdir_name}, skipping...")
|
120 |
+
return None
|
121 |
+
|
122 |
+
detections_data, questions_data, mask_file, source_file, (img_width, img_height) = file_data
|
123 |
+
|
124 |
+
# Create FiftyOne sample
|
125 |
+
sample = fo.Sample(filepath=str(source_file.absolute()))
|
126 |
+
|
127 |
+
# Add detections
|
128 |
+
sample['detections'] = convert_detections_to_relative(detections_data, img_width, img_height)
|
129 |
+
|
130 |
+
# Add segmentation mask
|
131 |
+
sample['segmentation'] = fo.Segmentation(mask_path=str(mask_file.absolute()))
|
132 |
+
|
133 |
+
# Filter to English questions only and preserve order
|
134 |
+
english_questions = [q for q in questions_data if q.get('q_lang') == 'en']
|
135 |
+
|
136 |
+
# Add sample-level metadata
|
137 |
+
add_sample_metadata(sample, english_questions)
|
138 |
+
|
139 |
+
# Add individual questions and answers
|
140 |
+
add_questions_and_answers(sample, english_questions)
|
141 |
+
|
142 |
+
return sample
|
143 |
+
|
144 |
+
def parse_slake_dataset(data_root="SLAKE/imgs", dataset_name="SLAKE"):
|
145 |
+
"""
|
146 |
+
Parse SLAKE dataset into FiftyOne format.
|
147 |
+
|
148 |
+
Args:
|
149 |
+
data_root (str): Path to the SLAKE/imgs directory
|
150 |
+
dataset_name (str): Name for the FiftyOne dataset
|
151 |
+
|
152 |
+
Returns:
|
153 |
+
fo.Dataset: FiftyOne dataset with parsed samples
|
154 |
+
"""
|
155 |
+
dataset = fo.Dataset(dataset_name, overwrite=True)
|
156 |
+
|
157 |
+
data_root = Path(data_root)
|
158 |
+
samples = []
|
159 |
+
|
160 |
+
# Process each subdirectory
|
161 |
+
for subdir in data_root.iterdir():
|
162 |
+
if not subdir.is_dir():
|
163 |
+
continue
|
164 |
+
|
165 |
+
print(f"Processing {subdir.name}...")
|
166 |
+
sample = process_single_sample(subdir)
|
167 |
+
|
168 |
+
if sample is not None:
|
169 |
+
samples.append(sample)
|
170 |
+
|
171 |
+
# Add all samples to dataset efficiently
|
172 |
+
dataset.add_samples(samples)
|
173 |
+
dataset.compute_metadata()
|
174 |
+
|
175 |
+
return dataset
|
176 |
+
|
177 |
+
import fiftyone as fo
|
178 |
+
from pathlib import Path
|
179 |
+
|
180 |
+
def load_mask_targets_from_file(mask_targets_file):
|
181 |
+
"""
|
182 |
+
Load mask targets mapping from file.
|
183 |
+
|
184 |
+
Args:
|
185 |
+
mask_targets_file (str): Path to the mask targets file
|
186 |
+
|
187 |
+
Returns:
|
188 |
+
dict: Mapping of pixel values to organ labels
|
189 |
+
"""
|
190 |
+
mask_targets = {}
|
191 |
+
|
192 |
+
with open(mask_targets_file, 'r') as f:
|
193 |
+
for line in f:
|
194 |
+
line = line.strip()
|
195 |
+
if ':' in line:
|
196 |
+
pixel_value, label = line.split(':', 1)
|
197 |
+
mask_targets[int(pixel_value)] = label
|
198 |
+
|
199 |
+
return mask_targets
|
200 |
+
|
201 |
+
def set_dataset_mask_targets(dataset_name, mask_targets_file, segmentation_field="segmentation"):
|
202 |
+
"""
|
203 |
+
Set mask targets for an existing FiftyOne dataset.
|
204 |
+
|
205 |
+
Args:
|
206 |
+
dataset_name (str): Name of the FiftyOne dataset
|
207 |
+
mask_targets_file (str): Path to the mask targets mapping file
|
208 |
+
segmentation_field (str): Name of the segmentation field (default: "segmentation")
|
209 |
+
"""
|
210 |
+
# Load dataset
|
211 |
+
dataset = fo.load_dataset(dataset_name)
|
212 |
+
|
213 |
+
# Load mask targets from file
|
214 |
+
mask_targets = load_mask_targets_from_file(mask_targets_file)
|
215 |
+
|
216 |
+
# Set mask targets
|
217 |
+
dataset.mask_targets = {segmentation_field: mask_targets}
|
218 |
+
dataset.save() # Must save after setting mask targets
|
219 |
+
|
220 |
+
for i, (pixel_val, label) in enumerate(list(mask_targets.items())[:5]):
|
221 |
+
print(f" {pixel_val}: {label}")
|
222 |
+
if len(mask_targets) > 5:
|
223 |
+
print(f" ... and {len(mask_targets) - 5} more")
|
224 |
+
|
225 |
+
|
226 |
+
|
227 |
+
dataset = parse_slake_dataset("SLAKE/imgs", "SLAKE")
|
228 |
+
|
229 |
+
set_dataset_mask_targets(
|
230 |
+
dataset_name="SLAKE", # Your dataset name
|
231 |
+
mask_targets_file="SLAKE/mask.txt", # Your mapping file
|
232 |
+
segmentation_field="segmentation"
|
233 |
+
)
|