Datasets:

Modalities:
Image
Text
Formats:
parquet
Languages:
English
DOI:
License:
mueller-franzes commited on
Commit
246798f
·
verified ·
1 Parent(s): e51c822

Update README.md

Browse files

fix typos and polish TAIX-Ray dataset README

Files changed (1) hide show
  1. README.md +42 -26
README.md CHANGED
@@ -127,36 +127,47 @@ dataset_info:
127
  dataset_size: 1226435862593.624
128
  ---
129
 
 
130
  # TAIX-Ray Dataset
131
 
132
- TAIX-Ray is a comprehensive dataset of about 200k bedside chest radiographs from about 50k intensive care patients at the University Hospital in Aachen, Germany, collected between 2010 and 2024.
133
- Trained radiologists provided structured reports at the time of acquisition, assessing key findings such as cardiomegaly, pulmonary congestion, pleural effusion, pulmonary opacities, and atelectasis on an ordinal scale.
134
 
 
135
 
136
  <br>
137
 
138
- ## Code & Details:
139
- The code for data loading, preprocessing, and baseline experiments is available at: https://github.com/mueller-franzes/TAIX-Ray
 
 
 
 
140
 
141
  ## How to Use
142
 
143
  ### Prerequisites
 
144
  Ensure you have the following dependencies installed:
145
 
146
  ```bash
147
  pip install datasets matplotlib huggingface_hub pandas tqdm
148
  ```
149
 
150
- ### Configurations
151
- This dataset is available in two configurations.
 
 
 
152
 
153
  | **Name** | **Size** | **Image Size** |
154
- |------------|----------|----------------|
155
- | default | 62GB | 512px |
156
- | original | 1.2TB | variable |
157
 
 
 
 
158
 
159
- ### Option A: Use within the Hugging Face Framework
160
  If you want to use the dataset directly within the Hugging Face `datasets` library, you can load and visualize it as follows:
161
 
162
  ```python
@@ -167,26 +178,29 @@ from matplotlib import pyplot as plt
167
  dataset = load_dataset("TLAIM/TAIX-Ray", name="default")
168
 
169
  # Access the training split (Fold 0)
170
- ds_train = dataset['train']
171
 
172
  # Retrieve a single sample from the training set
173
  item = ds_train[0]
174
 
175
  # Extract and display the image
176
- image = item['Image']
177
- plt.imshow(image, cmap='gray')
178
- plt.savefig('image.png') # Save the image to a file
179
  plt.show() # Display the image
180
 
181
  # Print metadata (excluding the image itself)
182
  for key in item.keys():
183
- if key != 'Image':
184
  print(f"{key}: {item[key]}")
185
  ```
186
 
187
- ### Option B: Downloading the Dataset
 
 
188
 
189
  If you prefer to download the dataset to a specific folder, use the following script. This will create the following folder structure:
 
190
  ```
191
  .
192
  ├── data/
@@ -194,13 +208,12 @@ If you prefer to download the dataset to a specific folder, use the following sc
194
  │ ├── d8546c6108aad271211da996eb7e9eeabaf44d39cf0226a4301c3cbe12d84151.png
195
  │ └── ...
196
  └── metadata/
197
- ├── annoation.csv
198
- └── split.csv
199
  ```
200
 
201
  ```python
202
  from datasets import load_dataset
203
- from huggingface_hub import hf_hub_download
204
  from pathlib import Path
205
  import pandas as pd
206
  from tqdm import tqdm
@@ -208,19 +221,19 @@ from tqdm import tqdm
208
  # Define output paths
209
  output_root = Path("./TAIX-Ray")
210
 
211
- # Create folders
212
  data_dir = output_root / "data"
213
  metadata_dir = output_root / "metadata"
214
  data_dir.mkdir(parents=True, exist_ok=True)
215
  metadata_dir.mkdir(parents=True, exist_ok=True)
216
 
217
  # Load dataset in streaming mode
218
- dataset = dataset = load_dataset("TLAIM/TAIX-Ray", name="default", streaming=True)
219
 
220
  # Process dataset
221
  metadata = []
222
  for split, split_dataset in dataset.items():
223
- print("-------- Start Download: ", split, " --------")
224
  for item in tqdm(split_dataset, desc="Downloading"): # Stream data one-by-one
225
  uid = item["UID"]
226
  img = item.pop("Image") # PIL Image object
@@ -229,16 +242,19 @@ for split, split_dataset in dataset.items():
229
  img.save(data_dir / f"{uid}.png", format="PNG")
230
 
231
  # Store metadata
232
- metadata.append(item)
233
 
234
  # Convert metadata to DataFrame
235
  metadata_df = pd.DataFrame(metadata)
236
 
237
- # Save annotations to CSV files
238
- metadata_df.drop(columns=["Split", "Fold"]).to_csv(metadata_dir / "annotation.csv", index=False)
239
-
 
240
 
241
  print("Dataset streamed and saved successfully!")
242
  ```
243
 
244
 
 
 
 
127
  dataset_size: 1226435862593.624
128
  ---
129
 
130
+
131
  # TAIX-Ray Dataset
132
 
133
+ TAIX-Ray is a comprehensive dataset of approximately 200k bedside chest radiographs from around 50k intensive care patients at University Hospital Aachen, Germany, collected between 2010 and 2024.
 
134
 
135
+ Trained radiologists provided structured reports at the time of acquisition, assessing key findings such as cardiomegaly, pulmonary congestion, pleural effusion, pulmonary opacities, and atelectasis on an ordinal scale.
136
 
137
  <br>
138
 
139
+ ## Code & Details
140
+
141
+ The code for data loading, preprocessing, and baseline experiments is available at:
142
+ [https://github.com/mueller-franzes/TAIX-Ray](https://github.com/mueller-franzes/TAIX-Ray)
143
+
144
+ ---
145
 
146
  ## How to Use
147
 
148
  ### Prerequisites
149
+
150
  Ensure you have the following dependencies installed:
151
 
152
  ```bash
153
  pip install datasets matplotlib huggingface_hub pandas tqdm
154
  ```
155
 
156
+ ---
157
+
158
+ ## Configurations
159
+
160
+ This dataset is available in two configurations:
161
 
162
  | **Name** | **Size** | **Image Size** |
163
+ | -------- | -------- | -------------- |
164
+ | default | 62GB | 512px |
165
+ | original | 1.2TB | variable |
166
 
167
+ ---
168
+
169
+ ## Option A: Use within the Hugging Face Framework
170
 
 
171
  If you want to use the dataset directly within the Hugging Face `datasets` library, you can load and visualize it as follows:
172
 
173
  ```python
 
178
  dataset = load_dataset("TLAIM/TAIX-Ray", name="default")
179
 
180
  # Access the training split (Fold 0)
181
+ ds_train = dataset["train"]
182
 
183
  # Retrieve a single sample from the training set
184
  item = ds_train[0]
185
 
186
  # Extract and display the image
187
+ image = item["Image"]
188
+ plt.imshow(image, cmap="gray")
189
+ plt.savefig("image.png") # Save the image to a file
190
  plt.show() # Display the image
191
 
192
  # Print metadata (excluding the image itself)
193
  for key in item.keys():
194
+ if key != "Image":
195
  print(f"{key}: {item[key]}")
196
  ```
197
 
198
+ ---
199
+
200
+ ## Option B: Downloading the Dataset
201
 
202
  If you prefer to download the dataset to a specific folder, use the following script. This will create the following folder structure:
203
+
204
  ```
205
  .
206
  ├── data/
 
208
  │ ├── d8546c6108aad271211da996eb7e9eeabaf44d39cf0226a4301c3cbe12d84151.png
209
  │ └── ...
210
  └── metadata/
211
+ ├── annotation.csv
212
+ └── split.csv
213
  ```
214
 
215
  ```python
216
  from datasets import load_dataset
 
217
  from pathlib import Path
218
  import pandas as pd
219
  from tqdm import tqdm
 
221
  # Define output paths
222
  output_root = Path("./TAIX-Ray")
223
 
224
+ # Create folders
225
  data_dir = output_root / "data"
226
  metadata_dir = output_root / "metadata"
227
  data_dir.mkdir(parents=True, exist_ok=True)
228
  metadata_dir.mkdir(parents=True, exist_ok=True)
229
 
230
  # Load dataset in streaming mode
231
+ dataset = load_dataset("TLAIM/TAIX-Ray", name="default", streaming=True)
232
 
233
  # Process dataset
234
  metadata = []
235
  for split, split_dataset in dataset.items():
236
+ print("-------- Start Download:", split, "--------")
237
  for item in tqdm(split_dataset, desc="Downloading"): # Stream data one-by-one
238
  uid = item["UID"]
239
  img = item.pop("Image") # PIL Image object
 
242
  img.save(data_dir / f"{uid}.png", format="PNG")
243
 
244
  # Store metadata
245
+ metadata.append(item)
246
 
247
  # Convert metadata to DataFrame
248
  metadata_df = pd.DataFrame(metadata)
249
 
250
+ # Save annotations to CSV file
251
+ metadata_df.drop(columns=["Split", "Fold"]).to_csv(
252
+ metadata_dir / "annotation.csv", index=False
253
+ )
254
 
255
  print("Dataset streamed and saved successfully!")
256
  ```
257
 
258
 
259
+
260
+