Datasets:
Iulia Elisa
commited on
Commit
·
49715f1
1
Parent(s):
6accdbd
minor changes
Browse files- README.md +2 -17
- load_and_visualise_dataset.ipynb +0 -0
- utils.py +0 -322
README.md
CHANGED
@@ -90,29 +90,14 @@ xami_dataset = XAMIDataset(
|
|
90 |
dataset_name="xami_dataset",
|
91 |
data_path='./dest_dir')
|
92 |
```
|
93 |
-
|
|
|
94 |
|
95 |
-
- using a CLI command
|
96 |
```bash
|
97 |
DEST_DIR='/path/to/local/dataset/dir'
|
98 |
|
99 |
huggingface-cli download iulia-elisa/XAMI-dataset xami_dataset.zip --repo-type dataset --local-dir "$DEST_DIR" && unzip "$DEST_DIR/xami_dataset.zip" -d "$DEST_DIR" && rm "$DEST_DIR/xami_dataset.zip"
|
100 |
-
|
101 |
```
|
102 |
-
<!--
|
103 |
-
# Dataset Split with SKF (Optional)
|
104 |
-
|
105 |
-
- The below method allows for dataset splitting, using the pre-generated splits in CSV files. This step is useful when training multiple dataset splits versions to gain mor generalised view on metrics.
|
106 |
-
```python
|
107 |
-
import utils
|
108 |
-
|
109 |
-
# run multilabel SKF split with the standard k=4
|
110 |
-
csv_files = ['mskf_0.csv', 'mskf_1.csv', 'mskf_2.csv', 'mskf_3.csv']
|
111 |
-
|
112 |
-
for idx, csv_file in enumerate(csv_files):
|
113 |
-
mskf = pd.read_csv(csv_file)
|
114 |
-
utils.create_directories_and_copy_files(images_dir, data_in, mskf, idx)
|
115 |
-
``` -->
|
116 |
|
117 |
## Licence
|
118 |
**[CC BY-NC 3.0 IGO](https://creativecommons.org/licenses/by-nc/3.0/igo/deed.en).**
|
|
|
90 |
dataset_name="xami_dataset",
|
91 |
data_path='./dest_dir')
|
92 |
```
|
93 |
+
###
|
94 |
+
Or you can simply download only the dataset zip file from HuggingFace using a CLI command:
|
95 |
|
|
|
96 |
```bash
|
97 |
DEST_DIR='/path/to/local/dataset/dir'
|
98 |
|
99 |
huggingface-cli download iulia-elisa/XAMI-dataset xami_dataset.zip --repo-type dataset --local-dir "$DEST_DIR" && unzip "$DEST_DIR/xami_dataset.zip" -d "$DEST_DIR" && rm "$DEST_DIR/xami_dataset.zip"
|
|
|
100 |
```
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
101 |
|
102 |
## Licence
|
103 |
**[CC BY-NC 3.0 IGO](https://creativecommons.org/licenses/by-nc/3.0/igo/deed.en).**
|
load_and_visualise_dataset.ipynb
DELETED
The diff for this file is too large to render.
See raw diff
|
|
utils.py
DELETED
@@ -1,322 +0,0 @@
|
|
1 |
-
import os
|
2 |
-
import json
|
3 |
-
from shutil import copy
|
4 |
-
import pandas as pd
|
5 |
-
from pathlib import Path
|
6 |
-
from PIL import Image, ImageDraw
|
7 |
-
import cv2
|
8 |
-
import numpy as np
|
9 |
-
import re
|
10 |
-
import datasets
|
11 |
-
from datasets import Value
|
12 |
-
from io import BytesIO
|
13 |
-
from PIL import Image, ImageDraw, ImageFont
|
14 |
-
import matplotlib.pyplot as plt
|
15 |
-
import matplotlib.patches as patches
|
16 |
-
|
17 |
-
def create_directories_and_copy_files(images_dir, coco_data, image_data, k):
|
18 |
-
base_dir = os.path.join(images_dir, f'mskf_{k}')
|
19 |
-
os.makedirs(base_dir, exist_ok=True)
|
20 |
-
|
21 |
-
for split in np.unique(image_data['SPLIT']):
|
22 |
-
split_dir = os.path.join(base_dir, split)
|
23 |
-
os.makedirs(split_dir, exist_ok=True)
|
24 |
-
|
25 |
-
# Filter the annotations
|
26 |
-
split_ids = image_data[image_data['SPLIT'] == split]['IMADE_ID'].tolist()
|
27 |
-
annotations = {
|
28 |
-
'images': [img for img in coco_data['images'] if img['id'] in split_ids],
|
29 |
-
'annotations': [ann for ann in coco_data['annotations'] if ann['image_id'] in split_ids],
|
30 |
-
'categories': coco_data['categories']
|
31 |
-
}
|
32 |
-
|
33 |
-
# Write the filtered annotations to a file
|
34 |
-
with open(os.path.join(split_dir, '_annotations.coco.json'), 'w') as f:
|
35 |
-
json.dump(annotations, f, indent=4)
|
36 |
-
|
37 |
-
# Copy the images
|
38 |
-
split_data = image_data[image_data['SPLIT'] == split]
|
39 |
-
for _, row in split_data.iterrows():
|
40 |
-
source = row['IMAGE_PATH']
|
41 |
-
destination = os.path.join(split_dir, os.path.basename(source))
|
42 |
-
copy(source, destination)
|
43 |
-
|
44 |
-
print(f'Dataset split for mskf_{k} was successful.')
|
45 |
-
|
46 |
-
def split_to_df(dataset_dir, split):
|
47 |
-
annotations_path = Path(dataset_dir+split+'/_annotations.coco.json')
|
48 |
-
|
49 |
-
with annotations_path.open('r') as f:
|
50 |
-
coco_data = json.load(f)
|
51 |
-
|
52 |
-
def image_from_path(file_path):
|
53 |
-
image = cv2.imread(file_path)
|
54 |
-
return image
|
55 |
-
|
56 |
-
def gen_segmentation(segmentation, width, height):
|
57 |
-
mask_img = np.zeros((height, width, 3), dtype=np.uint8)
|
58 |
-
for segment in segmentation:
|
59 |
-
pts = np.array(segment, np.int32).reshape((-1, 1, 2))
|
60 |
-
cv2.fillPoly(mask_img, [pts], (255, 255, 255)) # Fill color in BGR
|
61 |
-
|
62 |
-
return mask_img
|
63 |
-
|
64 |
-
images_df = pd.DataFrame(coco_data['images'][:50], columns=['id', 'file_name', 'width', 'height'])
|
65 |
-
annotations_df = pd.DataFrame(coco_data['annotations'])
|
66 |
-
df = pd.merge(annotations_df, images_df, left_on='image_id', right_on='id')
|
67 |
-
image_folder = annotations_path.parent
|
68 |
-
df['file_path'] = df['file_name'].apply(lambda x: str(image_folder / x))
|
69 |
-
df['observation'] = df['file_name'].apply(lambda x: x.split('.')[0].replace('_png', ''))
|
70 |
-
df['image'] = df['file_path'].apply(image_from_path)
|
71 |
-
df['segmentation'] = df.apply(lambda row: gen_segmentation(row['segmentation'], row['width'], row['height']), axis=1)
|
72 |
-
df = df.drop('file_path', axis=1)
|
73 |
-
df = df.drop('file_name', axis=1)
|
74 |
-
df['annot_id'] = df['id_x']
|
75 |
-
df = df.drop('id_x', axis=1)
|
76 |
-
df = df.drop('id_y', axis=1)
|
77 |
-
|
78 |
-
# take image fro df, and the corresponging annotations and plot them on image
|
79 |
-
# for i in range(5):
|
80 |
-
# img = df['image'][i]
|
81 |
-
# annot_id = df['annot_id'][i]
|
82 |
-
# # plot the image with the annotation using plt
|
83 |
-
# if img.dtype != np.uint8:
|
84 |
-
# img = img.astype(np.uint8)
|
85 |
-
# # plot
|
86 |
-
# segm_polygon = df['segmentation'][i]
|
87 |
-
# plt.imshow(segm_polygon)
|
88 |
-
# plt.axis('off')
|
89 |
-
# plt.show()
|
90 |
-
# plt.close()
|
91 |
-
|
92 |
-
return df, coco_data
|
93 |
-
|
94 |
-
def df_to_dataset_dict(df, coco_data, cats_to_colours):
|
95 |
-
|
96 |
-
def annot_on_image(annot_id, img_array, cat_id, annot_type='segm'):
|
97 |
-
if img_array.dtype != np.uint8:
|
98 |
-
img_array = img_array.astype(np.uint8)
|
99 |
-
|
100 |
-
pil_image = Image.fromarray(img_array)
|
101 |
-
draw = ImageDraw.Draw(pil_image)
|
102 |
-
if annot_type=='bbox':
|
103 |
-
bbox = [annot for annot in coco_data['annotations'] if annot['id'] == annot_id][0]['bbox']
|
104 |
-
x_min, y_min, width, height = bbox
|
105 |
-
top_left = (x_min, y_min)
|
106 |
-
bottom_right = (x_min + width, y_min + height)
|
107 |
-
|
108 |
-
draw.rectangle([top_left, bottom_right], outline=cats_to_colours[cat_id][1], width=2)
|
109 |
-
else:
|
110 |
-
# look for the annotation in coco_data that corresponds to the annot_id
|
111 |
-
segm_polygon = [annot for annot in coco_data['annotations'] if annot['id'] == annot_id][0]['segmentation'][0]
|
112 |
-
polygon = [(segm_polygon[i], segm_polygon[i+1]) for i in range(0, len(segm_polygon), 2)]
|
113 |
-
draw.polygon(polygon, outline=cats_to_colours[cat_id][1], width=2)
|
114 |
-
|
115 |
-
# plt.imshow(pil_image)
|
116 |
-
# plt.axis('off')
|
117 |
-
# plt.show()
|
118 |
-
# plt.close()
|
119 |
-
|
120 |
-
byte_io = BytesIO()
|
121 |
-
pil_image.save(byte_io, 'PNG')
|
122 |
-
byte_io.seek(0)
|
123 |
-
png_image = Image.open(byte_io)
|
124 |
-
|
125 |
-
return png_image
|
126 |
-
|
127 |
-
dictionary = df.to_dict(orient='list')
|
128 |
-
feats=datasets.Features({"observation id":Value(dtype='string'), \
|
129 |
-
'segmentation': datasets.Image(), \
|
130 |
-
'bbox':datasets.Image() , \
|
131 |
-
'label': Value(dtype='string'),\
|
132 |
-
'area':Value(dtype='string'),
|
133 |
-
'image shape':Value(dtype='string')})
|
134 |
-
|
135 |
-
dataset_data = {"observation id":dictionary['observation'], \
|
136 |
-
'segmentation': [annot_on_image(dictionary['annot_id'][i], dictionary['image'][i], dictionary['category_id'][i]) \
|
137 |
-
for i in range(len(dictionary['segmentation']))], \
|
138 |
-
'bbox': [annot_on_image(dictionary['annot_id'][i], dictionary['image'][i], dictionary['category_id'][i], annot_type='bbox') \
|
139 |
-
for i in range(len(dictionary['bbox']))], \
|
140 |
-
'label': [cats_to_colours[cat][0] for cat in dictionary['category_id']],\
|
141 |
-
'area':['%.3f'%(value) for value in dictionary['area']], \
|
142 |
-
'image shape':[f"({dictionary['width'][i]}, {dictionary['height'][i]})" for i in range(len(dictionary['width']))]}
|
143 |
-
the_dataset=datasets.Dataset.from_dict(dataset_data,features=feats)
|
144 |
-
|
145 |
-
return the_dataset
|
146 |
-
|
147 |
-
def merge_coco_jsons(first_json, second_json, output_path):
|
148 |
-
|
149 |
-
# Load the first JSON file
|
150 |
-
with open(first_json) as f:
|
151 |
-
coco1 = json.load(f)
|
152 |
-
|
153 |
-
# Load the second JSON file
|
154 |
-
with open(second_json) as f:
|
155 |
-
coco2 = json.load(f)
|
156 |
-
|
157 |
-
# Update IDs in coco2 to ensure they are unique and do not overlap with coco1
|
158 |
-
max_image_id = max(image['id'] for image in coco1['images'])
|
159 |
-
max_annotation_id = max(annotation['id'] for annotation in coco1['annotations'])
|
160 |
-
max_category_id = max(category['id'] for category in coco1['categories'])
|
161 |
-
|
162 |
-
# Add an offset to the second coco IDs
|
163 |
-
image_id_offset = max_image_id + 1
|
164 |
-
annotation_id_offset = max_annotation_id + 1
|
165 |
-
# category_id_offset = max_category_id + 1
|
166 |
-
|
167 |
-
# Apply offset to images, annotations, and categories in the second JSON
|
168 |
-
for image in coco2['images']:
|
169 |
-
image['id'] += image_id_offset
|
170 |
-
|
171 |
-
for annotation in coco2['annotations']:
|
172 |
-
annotation['id'] += annotation_id_offset
|
173 |
-
annotation['image_id'] += image_id_offset # Update the image_id reference
|
174 |
-
|
175 |
-
# Merge the two datasets
|
176 |
-
merged_coco = {
|
177 |
-
'images': coco1['images'] + coco2['images'],
|
178 |
-
'annotations': coco1['annotations'] + coco2['annotations'],
|
179 |
-
'categories': coco1['categories'] # If categories are the same; otherwise, merge as needed
|
180 |
-
}
|
181 |
-
|
182 |
-
# Save the merged annotations to a new JSON file
|
183 |
-
with open(output_path, 'w') as f:
|
184 |
-
json.dump(merged_coco, f)
|
185 |
-
|
186 |
-
def percentages(n_splits, image_ids, labels):
|
187 |
-
labels_percentages = {}
|
188 |
-
for i in range(n_splits):
|
189 |
-
train_k, valid_k = 0, 0
|
190 |
-
train_labels_counts = {'0':0, '1':0, '2':0, '3':0, '4':0, '5':0}
|
191 |
-
valid_labels_counts = {'0':0, '1':0, '2':0, '3':0, '4':0, '5':0}
|
192 |
-
for j in range(len(image_ids[i]['train'])):
|
193 |
-
for cat in list(labels[i]['train'][j]):
|
194 |
-
train_labels_counts[cat] += 1
|
195 |
-
train_k+=1
|
196 |
-
|
197 |
-
for j in range(len(image_ids[i]['valid'])):
|
198 |
-
for cat in list(labels[i]['valid'][j]):
|
199 |
-
valid_labels_counts[cat] += 1
|
200 |
-
valid_k+=1
|
201 |
-
|
202 |
-
train_labels_counts = {cat:counts * 1.0/train_k for cat, counts in train_labels_counts.items()}
|
203 |
-
valid_labels_counts = {cat:counts * 1.0/valid_k for cat, counts in valid_labels_counts.items()}
|
204 |
-
|
205 |
-
labels_percentages[i] = {'train':train_labels_counts, 'valid': valid_labels_counts}
|
206 |
-
|
207 |
-
return labels_percentages
|
208 |
-
|
209 |
-
def make_split(data_in, train_index, valid_index):
|
210 |
-
|
211 |
-
data_in_train = data_in.copy()
|
212 |
-
data_in_valid = data_in.copy()
|
213 |
-
|
214 |
-
data_in_train['images'] = [data_in['images'][train_index[i][0]] for i in range(len(train_index))]
|
215 |
-
data_in_valid['images'] = [data_in['images'][valid_index[i][0]] for i in range(len(valid_index))]
|
216 |
-
train_annot_ids, valid_annot_ids = [], []
|
217 |
-
|
218 |
-
for img_i in data_in_train['images']:
|
219 |
-
annotation_ids = [annot['id'] for annot in data_in_train['annotations'] if annot['image_id'] == img_i['id']]
|
220 |
-
train_annot_ids +=annotation_ids
|
221 |
-
|
222 |
-
for img_i in data_in_valid['images']:
|
223 |
-
annotation_ids = [annot['id'] for annot in data_in_valid['annotations'] if annot['image_id'] == img_i['id']]
|
224 |
-
valid_annot_ids +=annotation_ids
|
225 |
-
|
226 |
-
data_in_train['annotations'] = [data_in_train['annotations'][id] for id in train_annot_ids]
|
227 |
-
data_in_valid['annotations'] = [data_in_valid['annotations'][id] for id in valid_annot_ids]
|
228 |
-
|
229 |
-
print(len(data_in_train['images']), len(data_in_valid['images']))
|
230 |
-
return data_in_train, data_in_valid
|
231 |
-
|
232 |
-
def correct_bboxes(annotations):
|
233 |
-
for ann in annotations:
|
234 |
-
# If the segmentation is in polygon format (COCO polygon)
|
235 |
-
if isinstance(ann['segmentation'], list):
|
236 |
-
|
237 |
-
points = np.array(ann['segmentation']).reshape(-1, 2)
|
238 |
-
x_min, y_min = np.inf, np.inf
|
239 |
-
x_max, y_max = -np.inf, -np.inf
|
240 |
-
x_min = min(x_min, points[:, 0].min())
|
241 |
-
y_min = min(y_min, points[:, 1].min())
|
242 |
-
x_max = max(x_max, points[:, 0].max())
|
243 |
-
y_max = max(y_max, points[:, 1].max())
|
244 |
-
|
245 |
-
width = x_max - x_min
|
246 |
-
height = y_max - y_min
|
247 |
-
|
248 |
-
# The bbox in COCO format [x_min, y_min, width, height]
|
249 |
-
bbox = [x_min, y_min, width, height]
|
250 |
-
x, y, w, h = map(int, bbox)
|
251 |
-
ann['bbox'] = [x, y, w, h]
|
252 |
-
|
253 |
-
return annotations
|
254 |
-
|
255 |
-
def highlight_max(s):
|
256 |
-
is_max = s == s.max()
|
257 |
-
return ['background-color: yellow' if v else '' for v in is_max]
|
258 |
-
|
259 |
-
def highlight_max_str(s):
|
260 |
-
|
261 |
-
cats = []
|
262 |
-
for cat in s:
|
263 |
-
cats.append([float(match) for match in re.findall(r"[-+]?[0-9]*\.?[0-9]+", cat)][0])
|
264 |
-
|
265 |
-
is_max = cats == np.max(cats)
|
266 |
-
return ['background-color: yellow' if v else '' for v in is_max]
|
267 |
-
|
268 |
-
def read_yolo_annotations(annotation_file):
|
269 |
-
with open(annotation_file, 'r') as file:
|
270 |
-
lines = file.readlines()
|
271 |
-
|
272 |
-
annotations = []
|
273 |
-
for line in lines:
|
274 |
-
parts = line.strip().split()
|
275 |
-
class_id = int(parts[0])
|
276 |
-
points = list(map(float, parts[1:]))
|
277 |
-
annotations.append((class_id, points))
|
278 |
-
|
279 |
-
return annotations
|
280 |
-
|
281 |
-
def display_image_with_annotations(coco, cat_names, image_id):
|
282 |
-
img = coco.loadImgs(image_id)[0]
|
283 |
-
image_path = os.path.join('./mskf_0/train/', img['file_name'])
|
284 |
-
I = Image.open(image_path)
|
285 |
-
plt.imshow(I); plt.axis('off')
|
286 |
-
ann_ids = coco.getAnnIds(imgIds=img['id'], iscrowd=None)
|
287 |
-
anns = coco.loadAnns(ann_ids)
|
288 |
-
ax = plt.gca()
|
289 |
-
|
290 |
-
for ann in anns:
|
291 |
-
bbox = ann['bbox']
|
292 |
-
rect = patches.Rectangle((bbox[0], bbox[1]), bbox[2], bbox[3],
|
293 |
-
linewidth=2, edgecolor='b', facecolor='none')
|
294 |
-
ax.add_patch(rect)
|
295 |
-
ax.text(bbox[0], bbox[1] - 5, cat_names[ann['category_id']],
|
296 |
-
color='blue', fontsize=12, bbox=dict(facecolor='white', alpha=0.5))
|
297 |
-
|
298 |
-
plt.show()
|
299 |
-
|
300 |
-
def plot_segmentations(image_path, annotations, category_mapping):
|
301 |
-
image = Image.open(image_path)
|
302 |
-
width, height = image.size
|
303 |
-
draw = ImageDraw.Draw(image)
|
304 |
-
|
305 |
-
try:
|
306 |
-
font = ImageFont.truetype("DejaVuSans.ttf", 16) # Load a font
|
307 |
-
except IOError:
|
308 |
-
font = ImageFont.load_default()
|
309 |
-
|
310 |
-
for class_id, points in annotations:
|
311 |
-
# Scale points from normalized coordinates to image dimensions
|
312 |
-
scaled_points = [(p[0] * width, p[1] * height) for p in zip(points[0::2], points[1::2])]
|
313 |
-
draw.polygon(scaled_points, outline='green', fill=None)
|
314 |
-
|
315 |
-
category_name = category_mapping[class_id][0]
|
316 |
-
centroid_x = sum([p[0] for p in scaled_points]) / len(scaled_points)
|
317 |
-
centroid_y = sum([p[1] for p in scaled_points]) / len(scaled_points)
|
318 |
-
draw.text((centroid_x, centroid_y), category_name, fill='red', font=font, anchor='ms')
|
319 |
-
|
320 |
-
plt.imshow(image)
|
321 |
-
plt.axis('off')
|
322 |
-
plt.show()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|