✨ [Add] json can be directly read by dataloader.py
Browse filesThe default format is json in this version. I also done refactored it.
- utils/dataloader.py +98 -57
utils/dataloader.py
CHANGED
@@ -2,7 +2,7 @@ import json
|
|
2 |
import os
|
3 |
from itertools import chain
|
4 |
from os import listdir, path
|
5 |
-
from typing import List, Tuple, Union
|
6 |
|
7 |
import diskcache as dc
|
8 |
import hydra
|
@@ -17,7 +17,17 @@ from torchvision.transforms import functional as TF
|
|
17 |
from tqdm.rich import tqdm
|
18 |
|
19 |
|
20 |
-
def find_labels_path(dataset_path, phase_name):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
21 |
json_labels_path = path.join(dataset_path, "annotations", f"instances_{phase_name}.json")
|
22 |
|
23 |
txt_labels_path = path.join(dataset_path, "label", phase_name)
|
@@ -33,34 +43,74 @@ def find_labels_path(dataset_path, phase_name):
|
|
33 |
raise FileNotFoundError("No labels found in the specified dataset path and phase name.")
|
34 |
|
35 |
|
36 |
-
def
|
37 |
-
|
38 |
-
|
39 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
40 |
|
|
|
|
|
|
|
|
|
|
|
41 |
|
42 |
-
|
|
|
|
|
|
|
|
|
43 |
annotation_lookup = {}
|
44 |
for anno in data["annotations"]:
|
45 |
-
if anno["iscrowd"]
|
46 |
-
|
47 |
-
|
48 |
-
|
49 |
-
annotation_lookup[image_id]
|
|
|
50 |
return annotation_lookup
|
51 |
|
52 |
|
53 |
-
def
|
54 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
55 |
h, w = image_dimensions["height"], image_dimensions["width"]
|
56 |
for anno in annotations:
|
57 |
category_id = anno["category_id"]
|
58 |
-
|
59 |
-
|
60 |
-
|
61 |
-
|
62 |
-
|
63 |
-
|
|
|
|
|
64 |
|
65 |
|
66 |
class YoloDataset(Dataset):
|
@@ -114,48 +164,39 @@ class YoloDataset(Dataset):
|
|
114 |
images_path = path.join(dataset_path, "images", phase_name)
|
115 |
labels_path, data_type = find_labels_path(dataset_path, phase_name)
|
116 |
images_list = sorted(os.listdir(images_path))
|
117 |
-
data = []
|
118 |
-
valid_inputs = 0
|
119 |
-
|
120 |
if data_type == "json":
|
121 |
-
|
122 |
-
annotations_lookup = create_annotation_lookup(labels_data)
|
123 |
-
image_info_dict = {path.splitext(img["file_name"])[0]: img for img in labels_data["images"]}
|
124 |
|
125 |
-
|
126 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
127 |
continue
|
128 |
-
|
129 |
-
|
130 |
-
|
131 |
-
|
132 |
-
if annotations:
|
133 |
-
processed_data = process_annotations(annotations, image_info["id"], image_info)
|
134 |
-
if processed_data:
|
135 |
-
img_path = path.join(images_path, image_name)
|
136 |
-
labels = self.load_valid_labels(img_path, processed_data)
|
137 |
-
if labels is not None:
|
138 |
-
data.append((img_path, labels))
|
139 |
-
valid_inputs += 1
|
140 |
-
|
141 |
-
elif data_type == "txt":
|
142 |
-
for image_name in tqdm(images_list, desc="Filtering data"):
|
143 |
-
if not image_name.lower().endswith((".jpg", ".jpeg", ".png")):
|
144 |
continue
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
145 |
img_path = path.join(images_path, image_name)
|
146 |
-
|
147 |
-
|
148 |
-
|
149 |
-
if path.isfile(label_path):
|
150 |
-
seg_data_one_img = []
|
151 |
-
with open(label_path, "r") as file:
|
152 |
-
for line in file:
|
153 |
-
parts = list(map(float, line.strip().split()))
|
154 |
-
seg_data_one_img.append(parts)
|
155 |
-
labels = self.load_valid_labels(label_path, seg_data_one_img)
|
156 |
-
if labels is not None:
|
157 |
-
data.append((img_path, labels))
|
158 |
-
valid_inputs += 1
|
159 |
|
160 |
logger.info("Recorded {}/{} valid inputs", valid_inputs, len(images_list))
|
161 |
return data
|
|
|
2 |
import os
|
3 |
from itertools import chain
|
4 |
from os import listdir, path
|
5 |
+
from typing import Any, Dict, List, Optional, Tuple, Union
|
6 |
|
7 |
import diskcache as dc
|
8 |
import hydra
|
|
|
17 |
from tqdm.rich import tqdm
|
18 |
|
19 |
|
20 |
+
def find_labels_path(dataset_path: str, phase_name: str):
|
21 |
+
"""
|
22 |
+
Find the path to label files for a specified dataset and phase(e.g. training).
|
23 |
+
|
24 |
+
Args:
|
25 |
+
dataset_path (str): The path to the root directory of the dataset.
|
26 |
+
phase_name (str): The name of the phase for which labels are being searched (e.g., "train", "val", "test").
|
27 |
+
|
28 |
+
Returns:
|
29 |
+
Tuple[str, str]: A tuple containing the path to the labels file and the file format ("json" or "txt").
|
30 |
+
"""
|
31 |
json_labels_path = path.join(dataset_path, "annotations", f"instances_{phase_name}.json")
|
32 |
|
33 |
txt_labels_path = path.join(dataset_path, "label", phase_name)
|
|
|
43 |
raise FileNotFoundError("No labels found in the specified dataset path and phase name.")
|
44 |
|
45 |
|
46 |
+
def create_image_info_dict(labels_path: str) -> Tuple[Dict[str, List], Dict[str, Dict]]:
|
47 |
+
"""
|
48 |
+
Create a dictionary containing image information and annotations indexed by image ID.
|
49 |
+
|
50 |
+
Args:
|
51 |
+
labels_path (str): The path to the annotation json file.
|
52 |
+
|
53 |
+
Returns:
|
54 |
+
- annotations_index: A dictionary where keys are image IDs and values are lists of annotations.
|
55 |
+
- image_info_dict: A dictionary where keys are image file names without extension and values are image information dictionaries.
|
56 |
+
"""
|
57 |
+
with open(labels_path, "r") as file:
|
58 |
+
labels_data = json.load(file)
|
59 |
+
annotations_index = index_annotations_by_image(labels_data) # check lookup is a good name?
|
60 |
+
image_info_dict = {path.splitext(img["file_name"])[0]: img for img in labels_data["images"]}
|
61 |
+
return annotations_index, image_info_dict
|
62 |
+
|
63 |
|
64 |
+
def index_annotations_by_image(data: Dict[str, Any]):
|
65 |
+
"""
|
66 |
+
Use image index to lookup every annotations
|
67 |
+
Args:
|
68 |
+
data (Dict[str, Any]): A dictionary containing annotation data.
|
69 |
|
70 |
+
Returns:
|
71 |
+
Dict[int, List[Dict[str, Any]]]: A dictionary where keys are image IDs and values are lists of annotations.
|
72 |
+
Annotations with "iscrowd" set to True are excluded from the index.
|
73 |
+
|
74 |
+
"""
|
75 |
annotation_lookup = {}
|
76 |
for anno in data["annotations"]:
|
77 |
+
if anno["iscrowd"]:
|
78 |
+
continue
|
79 |
+
image_id = anno["image_id"]
|
80 |
+
if image_id not in annotation_lookup:
|
81 |
+
annotation_lookup[image_id] = []
|
82 |
+
annotation_lookup[image_id].append(anno)
|
83 |
return annotation_lookup
|
84 |
|
85 |
|
86 |
+
def get_scaled_segmentation(
|
87 |
+
annotations: List[Dict[str, Any]], image_dimensions: Dict[str, int]
|
88 |
+
) -> Optional[List[List[float]]]:
|
89 |
+
"""
|
90 |
+
Scale the segmentation data based on image dimensions and return a list of scaled segmentation data.
|
91 |
+
|
92 |
+
Args:
|
93 |
+
annotations (List[Dict[str, Any]]): A list of annotation dictionaries.
|
94 |
+
image_dimensions (Dict[str, int]): A dictionary containing image dimensions (height and width).
|
95 |
+
|
96 |
+
Returns:
|
97 |
+
Optional[List[List[float]]]: A list of scaled segmentation data, where each sublist contains category_id followed by scaled (x, y) coordinates.
|
98 |
+
"""
|
99 |
+
if annotations is None:
|
100 |
+
return None
|
101 |
+
|
102 |
+
seg_array_with_cat = []
|
103 |
h, w = image_dimensions["height"], image_dimensions["width"]
|
104 |
for anno in annotations:
|
105 |
category_id = anno["category_id"]
|
106 |
+
seg_list = [item for sublist in anno["segmentation"] for item in sublist]
|
107 |
+
scaled_seg_data = (
|
108 |
+
np.array(seg_list).reshape(-1, 2) / [w, h]
|
109 |
+
).tolist() # make the list group in x, y pairs and scaled with image width, height
|
110 |
+
scaled_flat_seg_data = [category_id] + list(chain(*scaled_seg_data)) # flatten the scaled_seg_data list
|
111 |
+
seg_array_with_cat.append(scaled_flat_seg_data)
|
112 |
+
|
113 |
+
return seg_array_with_cat
|
114 |
|
115 |
|
116 |
class YoloDataset(Dataset):
|
|
|
164 |
images_path = path.join(dataset_path, "images", phase_name)
|
165 |
labels_path, data_type = find_labels_path(dataset_path, phase_name)
|
166 |
images_list = sorted(os.listdir(images_path))
|
|
|
|
|
|
|
167 |
if data_type == "json":
|
168 |
+
annotations_index, image_info_dict = create_image_info_dict(labels_path)
|
|
|
|
|
169 |
|
170 |
+
data = []
|
171 |
+
valid_inputs = 0
|
172 |
+
for image_name in tqdm(images_list, desc="Filtering data"):
|
173 |
+
if not image_name.lower().endswith((".jpg", ".jpeg", ".png")):
|
174 |
+
continue
|
175 |
+
image_id, _ = path.splitext(image_name)
|
176 |
+
|
177 |
+
if data_type == "json":
|
178 |
+
image_info = image_info_dict.get(image_id, None)
|
179 |
+
if image_info is None:
|
180 |
+
continue
|
181 |
+
annotations = annotations_index.get(image_info["id"], [])
|
182 |
+
image_seg_annotations = get_scaled_segmentation(annotations, image_info)
|
183 |
+
if not image_seg_annotations:
|
184 |
continue
|
185 |
+
|
186 |
+
elif data_type == "txt":
|
187 |
+
label_path = path.join(labels_path, f"{image_id}.txt")
|
188 |
+
if not path.isfile(label_path):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
189 |
continue
|
190 |
+
with open(label_path, "r") as file:
|
191 |
+
image_seg_annotations = [
|
192 |
+
list(map(float, line.strip().split())) for line in file
|
193 |
+
] # add a comment for this line, complicated, do you need "list", im not sure
|
194 |
+
|
195 |
+
labels = self.load_valid_labels(image_id, image_seg_annotations)
|
196 |
+
if labels is not None:
|
197 |
img_path = path.join(images_path, image_name)
|
198 |
+
data.append((img_path, labels))
|
199 |
+
valid_inputs += 1
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
200 |
|
201 |
logger.info("Recorded {}/{} valid inputs", valid_inputs, len(images_list))
|
202 |
return data
|