Upload paligemma_dataset.py with huggingface_hub
Browse files- paligemma_dataset.py +143 -175
paligemma_dataset.py
CHANGED
@@ -1,16 +1,14 @@
|
|
1 |
import json
|
2 |
import os
|
3 |
-
from pathlib import Path
|
4 |
-
|
5 |
import datasets
|
6 |
-
|
7 |
import jsonlines
|
8 |
|
9 |
logger = datasets.logging.get_logger(__name__)
|
10 |
|
11 |
_CITATION = """\
|
12 |
-
@misc{
|
13 |
-
title={
|
14 |
author={Chen, Xingqiang},
|
15 |
year={2024},
|
16 |
publisher={Hugging Face}
|
@@ -18,16 +16,18 @@ _CITATION = """\
|
|
18 |
"""
|
19 |
|
20 |
_DESCRIPTION = """\
|
21 |
-
|
22 |
-
|
|
|
|
|
23 |
The dataset includes both basic defect detection samples and a larger set of
|
24 |
-
874 annotated images from real-world structural inspections.
|
25 |
"""
|
26 |
|
27 |
-
_HOMEPAGE = "https://huggingface.co/datasets/xingqiang/
|
28 |
|
29 |
class PaligemmaDataset(datasets.GeneratorBasedBuilder):
|
30 |
-
"""
|
31 |
|
32 |
VERSION = datasets.Version("1.1.0")
|
33 |
|
@@ -39,7 +39,6 @@ class PaligemmaDataset(datasets.GeneratorBasedBuilder):
|
|
39 |
"boxes": datasets.Sequence(datasets.Sequence(datasets.Value("float32"), length=4)),
|
40 |
"labels": datasets.Sequence(datasets.ClassLabel(names=["void", "crack"])),
|
41 |
"caption": datasets.Value("string"),
|
42 |
-
"source": datasets.Value("string"),
|
43 |
}),
|
44 |
supervised_keys=None,
|
45 |
homepage=_HOMEPAGE,
|
@@ -70,121 +69,54 @@ class PaligemmaDataset(datasets.GeneratorBasedBuilder):
|
|
70 |
]
|
71 |
|
72 |
def _generate_examples(self, split):
|
73 |
-
"""
|
74 |
-
#
|
75 |
-
|
76 |
-
if os.path.exists(unified_path):
|
77 |
-
logger.info(f"使用统一格式的注释文件: {unified_path}")
|
78 |
-
with open(unified_path, encoding="utf-8") as f:
|
79 |
-
annotations = json.load(f)
|
80 |
-
for idx, ann in enumerate(annotations):
|
81 |
-
image_path = os.path.join("images", split, ann["image_filename"])
|
82 |
-
try:
|
83 |
-
yield f"unified_{idx}", {
|
84 |
-
"image": image_path,
|
85 |
-
"boxes": ann["boxes"],
|
86 |
-
"labels": ann["labels"],
|
87 |
-
"caption": ann["caption"],
|
88 |
-
"source": ann.get("source", "unified")
|
89 |
-
}
|
90 |
-
except Exception as e:
|
91 |
-
logger.warning(f"跳过无效图像 {image_path}: {e}")
|
92 |
-
continue
|
93 |
-
return # 如果使用了统一格式,不再处理其他格式
|
94 |
|
95 |
-
|
96 |
-
|
97 |
-
|
98 |
-
|
99 |
-
|
100 |
-
|
101 |
-
|
102 |
-
|
103 |
-
|
104 |
-
|
105 |
-
|
106 |
-
|
107 |
-
|
108 |
-
|
109 |
-
|
110 |
-
|
111 |
-
|
112 |
-
|
113 |
-
|
114 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
115 |
|
116 |
-
# 加载 JSONL 注释
|
117 |
-
jsonl_path = os.path.join("annotations", f"_annotations.{split}.jsonl")
|
118 |
-
if os.path.exists(jsonl_path):
|
119 |
-
with jsonlines.open(jsonl_path) as reader:
|
120 |
-
for idx, ann in enumerate(reader):
|
121 |
-
# 确保使用正确的图像文件名
|
122 |
-
image_filename = ann.get("image", "")
|
123 |
-
image_path = os.path.join("images", split, image_filename)
|
124 |
-
|
125 |
-
try:
|
126 |
-
# 不要尝试在这里打开图像,只返回路径
|
127 |
-
# 转换注释为我们的格式
|
128 |
-
if "annotations" in ann:
|
129 |
-
# 处理新格式
|
130 |
-
boxes = [[b["x"], b["y"], b["width"], b["height"]] for b in ann["annotations"]]
|
131 |
-
labels = [0 if b["class"] == "void" else 1 for b in ann["annotations"]]
|
132 |
-
caption = f"Image contains {len(boxes)} defects: " + \
|
133 |
-
", ".join([b["class"] for b in ann["annotations"]])
|
134 |
-
else:
|
135 |
-
# 处理旧格式 (prefix/suffix)
|
136 |
-
# 这里需要解析 suffix 中的位置信息
|
137 |
-
boxes = []
|
138 |
-
labels = []
|
139 |
-
if "suffix" in ann:
|
140 |
-
parts = ann["suffix"].split(";")
|
141 |
-
for part in parts:
|
142 |
-
part = part.strip()
|
143 |
-
if "<loc" in part:
|
144 |
-
# 解析位置和标签
|
145 |
-
loc_parts = part.split()
|
146 |
-
if len(loc_parts) >= 2:
|
147 |
-
# 提取坐标
|
148 |
-
coords = []
|
149 |
-
for loc in loc_parts[0].split("><"):
|
150 |
-
if loc.startswith("<loc"):
|
151 |
-
coords.append(int(loc[4:-1]) / 1024) # 归一化坐标
|
152 |
-
|
153 |
-
if len(coords) == 4:
|
154 |
-
boxes.append(coords)
|
155 |
-
label = 0 if "void" in loc_parts[1] else 1
|
156 |
-
labels.append(label)
|
157 |
-
|
158 |
-
caption = ann.get("prefix", "")
|
159 |
-
|
160 |
-
# 检查图像是否存在
|
161 |
-
image_exists = False
|
162 |
-
|
163 |
-
# 检查images/datasets目录
|
164 |
-
if os.path.exists(f"images/datasets/{image_filename}"):
|
165 |
-
image_exists = True
|
166 |
-
|
167 |
-
# 检查原始路径
|
168 |
-
if not image_exists:
|
169 |
-
for img_split in ["train", "val", "test"]:
|
170 |
-
if os.path.exists(f"images/{img_split}/{image_filename}"):
|
171 |
-
image_exists = True
|
172 |
-
break
|
173 |
-
|
174 |
-
if not image_exists:
|
175 |
-
print(f"警告: 图像文件不存在: {image_filename}")
|
176 |
-
continue
|
177 |
-
|
178 |
-
yield f"p1v1_{idx}", {
|
179 |
-
"image": image_path,
|
180 |
-
"boxes": boxes,
|
181 |
-
"labels": labels,
|
182 |
-
"caption": caption,
|
183 |
-
"source": "p1v1"
|
184 |
-
}
|
185 |
-
except Exception as e:
|
186 |
-
logger.warning(f"跳过��效注释 {image_path}: {e}")
|
187 |
-
continue
|
188 |
|
189 |
def convert_annotations_to_unified_format():
|
190 |
"""将所有注释转换为统一格式"""
|
@@ -220,21 +152,29 @@ def convert_annotations_to_unified_format():
|
|
220 |
else:
|
221 |
print(f"未找到 JSON 文件: {json_path}")
|
222 |
|
223 |
-
#
|
224 |
-
|
|
|
225 |
f"_annotations.{split}.jsonl",
|
226 |
-
f"_annotations.{split}1.jsonl"
|
227 |
-
f"p-1.v1i.paligemma/_annotations.{split}.jsonl",
|
228 |
-
f"p-1.v1i.paligemma/_annotations.{split}1.jsonl"
|
229 |
]
|
230 |
|
231 |
-
|
232 |
-
|
233 |
-
|
234 |
-
|
235 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
236 |
annotation_count = 0
|
237 |
-
with open(
|
238 |
for line_num, line in enumerate(f, 1):
|
239 |
try:
|
240 |
line = line.strip()
|
@@ -248,20 +188,24 @@ def convert_annotations_to_unified_format():
|
|
248 |
if not image_filename:
|
249 |
print(f"跳过第 {line_num} 行: 没有图像文件名")
|
250 |
continue
|
|
|
|
|
|
|
251 |
|
252 |
# 检查图像是否存在
|
253 |
image_exists = False
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
254 |
|
255 |
-
|
256 |
-
|
257 |
-
|
258 |
-
|
259 |
-
# 检查原始路径
|
260 |
-
if not image_exists:
|
261 |
-
for img_split in ["train", "val", "test"]:
|
262 |
-
if os.path.exists(f"images/{img_split}/{image_filename}"):
|
263 |
-
image_exists = True
|
264 |
-
break
|
265 |
|
266 |
if not image_exists:
|
267 |
print(f"警告: 图像文件不存在: {image_filename}")
|
@@ -281,26 +225,39 @@ def convert_annotations_to_unified_format():
|
|
281 |
caption = ann.get("prefix", "")
|
282 |
|
283 |
if "suffix" in ann:
|
284 |
-
parts = ann["suffix"].split(
|
285 |
-
for part in parts:
|
286 |
-
part = part.strip()
|
287 |
if "<loc" in part:
|
288 |
-
#
|
289 |
-
|
290 |
-
|
291 |
-
|
292 |
-
|
293 |
-
|
294 |
-
|
295 |
-
|
296 |
-
|
297 |
-
|
298 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
299 |
|
300 |
-
|
301 |
-
|
302 |
-
|
303 |
-
labels.append(label)
|
304 |
|
305 |
unified_annotations.append({
|
306 |
"image_filename": image_filename,
|
@@ -311,22 +268,33 @@ def convert_annotations_to_unified_format():
|
|
311 |
})
|
312 |
annotation_count += 1
|
313 |
except json.JSONDecodeError as e:
|
314 |
-
print(f"警告: {
|
315 |
continue
|
316 |
-
print(f"从 {
|
317 |
else:
|
318 |
-
print(f"未找到 JSONL 文件: {
|
319 |
|
320 |
# 如果是valid分割,与val合并
|
321 |
-
if split == "valid"
|
322 |
-
|
323 |
-
|
324 |
-
|
325 |
-
|
326 |
-
|
327 |
-
|
328 |
-
|
329 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
330 |
# 保存统一格式的注释
|
331 |
if unified_annotations:
|
332 |
# 对于valid分割,保存为val_unified.json
|
|
|
1 |
import json
|
2 |
import os
|
|
|
|
|
3 |
import datasets
|
4 |
+
|
5 |
import jsonlines
|
6 |
|
7 |
logger = datasets.logging.get_logger(__name__)
|
8 |
|
9 |
_CITATION = """\
|
10 |
+
@misc{chen2024gpradar,
|
11 |
+
title={GPRadar-Defect-MultiTask Dataset},
|
12 |
author={Chen, Xingqiang},
|
13 |
year={2024},
|
14 |
publisher={Hugging Face}
|
|
|
16 |
"""
|
17 |
|
18 |
_DESCRIPTION = """\
|
19 |
+
GPRadar-Defect-MultiTask Dataset
|
20 |
+
|
21 |
+
This dataset contains ground penetrating radar (GPR) images and annotations for defect detection and analysis,
|
22 |
+
designed for training and evaluating multimodal models for GPR defect detection.
|
23 |
The dataset includes both basic defect detection samples and a larger set of
|
24 |
+
874 annotated images from real-world structural inspections focusing on voids and cracks.
|
25 |
"""
|
26 |
|
27 |
+
_HOMEPAGE = "https://huggingface.co/datasets/xingqiang/GPRadar-Defect-MultiTask"
|
28 |
|
29 |
class PaligemmaDataset(datasets.GeneratorBasedBuilder):
|
30 |
+
"""GPRadar-Defect-MultiTask Dataset for GPR defect detection and analysis."""
|
31 |
|
32 |
VERSION = datasets.Version("1.1.0")
|
33 |
|
|
|
39 |
"boxes": datasets.Sequence(datasets.Sequence(datasets.Value("float32"), length=4)),
|
40 |
"labels": datasets.Sequence(datasets.ClassLabel(names=["void", "crack"])),
|
41 |
"caption": datasets.Value("string"),
|
|
|
42 |
}),
|
43 |
supervised_keys=None,
|
44 |
homepage=_HOMEPAGE,
|
|
|
69 |
]
|
70 |
|
71 |
def _generate_examples(self, split):
|
72 |
+
"""Yields examples."""
|
73 |
+
# 统一格式的注释文件
|
74 |
+
annotation_file = f"annotations/{split}_unified.json"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
75 |
|
76 |
+
if not os.path.exists(annotation_file):
|
77 |
+
# 如果统一格式文件不存在,尝试转换
|
78 |
+
convert_annotations_to_unified_format()
|
79 |
+
|
80 |
+
# 再次检查文件是否已创建
|
81 |
+
if not os.path.exists(annotation_file):
|
82 |
+
logger.warning(f"找不到统一格式注释文件: {annotation_file},将返回空数据")
|
83 |
+
return
|
84 |
+
|
85 |
+
# 加载统一格式的注释
|
86 |
+
with open(annotation_file, "r", encoding="utf-8") as f:
|
87 |
+
annotations = json.load(f)
|
88 |
+
|
89 |
+
for idx, ann in enumerate(annotations):
|
90 |
+
# 尝试在不同的可能路径中查找图���
|
91 |
+
image_found = False
|
92 |
+
image_filename = ann["image_filename"]
|
93 |
+
|
94 |
+
for image_path in [
|
95 |
+
f"images/{split}/{image_filename}",
|
96 |
+
f"images/datasets/{image_filename}",
|
97 |
+
f"images/{image_filename}",
|
98 |
+
]:
|
99 |
+
if os.path.exists(image_path):
|
100 |
+
yield idx, {
|
101 |
+
"image": image_path,
|
102 |
+
"boxes": ann["boxes"],
|
103 |
+
"labels": ann["labels"],
|
104 |
+
"caption": ann["caption"],
|
105 |
+
}
|
106 |
+
image_found = True
|
107 |
+
break
|
108 |
+
|
109 |
+
if not image_found:
|
110 |
+
logger.warning(f"找不到图像文件: {image_filename},跳过该示例")
|
111 |
+
|
112 |
+
|
113 |
+
def normalize_image_path(image_path):
|
114 |
+
"""规范化图像路径,移除多余的前缀"""
|
115 |
+
# 处理特殊前缀
|
116 |
+
if "p-1.v1i.paligemma-multimodal/dataset/" in image_path:
|
117 |
+
return image_path.split("p-1.v1i.paligemma-multimodal/dataset/")[-1]
|
118 |
+
return image_path
|
119 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
120 |
|
121 |
def convert_annotations_to_unified_format():
|
122 |
"""将所有注释转换为统一格式"""
|
|
|
152 |
else:
|
153 |
print(f"未找到 JSON 文件: {json_path}")
|
154 |
|
155 |
+
# 查找所有可能的JSONL注释文件
|
156 |
+
# 1. 检查根目录
|
157 |
+
jsonl_files_to_check = [
|
158 |
f"_annotations.{split}.jsonl",
|
159 |
+
f"_annotations.{split}1.jsonl"
|
|
|
|
|
160 |
]
|
161 |
|
162 |
+
# 2. 递归查找子目录中的JSONL文件
|
163 |
+
for root, dirs, files in os.walk("annotations"):
|
164 |
+
for file in files:
|
165 |
+
if file.endswith(f"{split}.jsonl") or file.endswith(f"{split}1.jsonl") or file.endswith(f"{split}2.jsonl"):
|
166 |
+
rel_path = os.path.relpath(os.path.join(root, file), "annotations")
|
167 |
+
if rel_path != file: # 不是根目录的文件
|
168 |
+
jsonl_files_to_check.append(rel_path)
|
169 |
+
|
170 |
+
# 处理所有找到的JSONL文件
|
171 |
+
for jsonl_path in jsonl_files_to_check:
|
172 |
+
full_path = os.path.join("annotations", jsonl_path)
|
173 |
+
print(f"检查 JSONL 文件: {full_path}")
|
174 |
+
if os.path.exists(full_path):
|
175 |
+
print(f"找到 JSONL 文件: {full_path}")
|
176 |
annotation_count = 0
|
177 |
+
with open(full_path, encoding="utf-8") as f:
|
178 |
for line_num, line in enumerate(f, 1):
|
179 |
try:
|
180 |
line = line.strip()
|
|
|
188 |
if not image_filename:
|
189 |
print(f"跳过第 {line_num} 行: 没有图像文件名")
|
190 |
continue
|
191 |
+
|
192 |
+
# 规范化图像路径
|
193 |
+
image_filename = normalize_image_path(image_filename)
|
194 |
|
195 |
# 检查图像是否存在
|
196 |
image_exists = False
|
197 |
+
possible_image_paths = [
|
198 |
+
f"images/datasets/{image_filename}",
|
199 |
+
f"images/train/{image_filename}",
|
200 |
+
f"images/val/{image_filename}",
|
201 |
+
f"images/test/{image_filename}",
|
202 |
+
f"images/{image_filename}" # 直接在images目录下
|
203 |
+
]
|
204 |
|
205 |
+
for img_path in possible_image_paths:
|
206 |
+
if os.path.exists(img_path):
|
207 |
+
image_exists = True
|
208 |
+
break
|
|
|
|
|
|
|
|
|
|
|
|
|
209 |
|
210 |
if not image_exists:
|
211 |
print(f"警告: 图像文件不存在: {image_filename}")
|
|
|
225 |
caption = ann.get("prefix", "")
|
226 |
|
227 |
if "suffix" in ann:
|
228 |
+
parts = ann["suffix"].split()
|
229 |
+
for i, part in enumerate(parts):
|
|
|
230 |
if "<loc" in part:
|
231 |
+
# 解析位置
|
232 |
+
coords = []
|
233 |
+
loc_str = part
|
234 |
+
while loc_str.startswith("<loc") and len(coords) < 4:
|
235 |
+
try:
|
236 |
+
# 提取坐标
|
237 |
+
coord_value = int(loc_str[4:loc_str.find(">")])
|
238 |
+
coords.append(coord_value / 1024) # 归一化坐标
|
239 |
+
# 移除已处理的部分
|
240 |
+
loc_str = loc_str[loc_str.find(">")+1:]
|
241 |
+
except (ValueError, IndexError):
|
242 |
+
break
|
243 |
+
|
244 |
+
if len(coords) == 4:
|
245 |
+
boxes.append(coords)
|
246 |
+
# 查找标签(通常在下一个部分)
|
247 |
+
label_idx = 1
|
248 |
+
while i + label_idx < len(parts) and not parts[i + label_idx].startswith("<loc"):
|
249 |
+
label_text = parts[i + label_idx]
|
250 |
+
if "void" in label_text:
|
251 |
+
labels.append(0)
|
252 |
+
break
|
253 |
+
elif "crack" in label_text:
|
254 |
+
labels.append(1)
|
255 |
+
break
|
256 |
+
label_idx += 1
|
257 |
|
258 |
+
# 如果未找到特定标签,默认为void
|
259 |
+
if len(labels) < len(boxes):
|
260 |
+
labels.append(0)
|
|
|
261 |
|
262 |
unified_annotations.append({
|
263 |
"image_filename": image_filename,
|
|
|
268 |
})
|
269 |
annotation_count += 1
|
270 |
except json.JSONDecodeError as e:
|
271 |
+
print(f"警告: {full_path} 第 {line_num} 行不是有效的 JSON: {e}")
|
272 |
continue
|
273 |
+
print(f"从 {full_path} 加载了 {annotation_count} 条注释")
|
274 |
else:
|
275 |
+
print(f"未找到 JSONL 文件: {full_path}")
|
276 |
|
277 |
# 如果是valid分割,与val合并
|
278 |
+
if split == "valid":
|
279 |
+
val_annotations = []
|
280 |
+
if os.path.exists(f"annotations/val_unified.json"):
|
281 |
+
try:
|
282 |
+
with open(f"annotations/val_unified.json", "r", encoding="utf-8") as f:
|
283 |
+
val_annotations = json.load(f)
|
284 |
+
print(f"加载现有val分割注释,共 {len(val_annotations)} 条记录")
|
285 |
+
|
286 |
+
# 合并注释,避免重复
|
287 |
+
existing_filenames = {ann["image_filename"] for ann in val_annotations}
|
288 |
+
for ann in unified_annotations:
|
289 |
+
if ann["image_filename"] not in existing_filenames:
|
290 |
+
val_annotations.append(ann)
|
291 |
+
existing_filenames.add(ann["image_filename"])
|
292 |
+
|
293 |
+
print(f"将valid分割与val分割合并,共 {len(val_annotations)} 条记录")
|
294 |
+
unified_annotations = val_annotations
|
295 |
+
except Exception as e:
|
296 |
+
print(f"合并valid和val分割时出错: {e}")
|
297 |
+
|
298 |
# 保存统一格式的注释
|
299 |
if unified_annotations:
|
300 |
# 对于valid分割,保存为val_unified.json
|