File size: 9,022 Bytes
b0c0df0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Contribute Your Models and Datasets\n",
    "\n",
    "[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/EvolvingLMMs-Lab/lmms-eval/blob/main/tools/make_image_hf_dataset.ipynb)\n",
    "\n",
    "This notebook will guide you to make correct format of Huggingface dataset, in proper parquet format and visualizable in Huggingface dataset hub.\n",
    "\n",
    "We will take the example of the dataset [`lmms-lab/VQAv2_TOY`](https://huggingface.co/datasets/lmms-lab/VQAv2_TOY) and convert it to the proper format."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Preparation\n",
    "\n",
    "We need to install `datasets` library to create the dataset and `Pillow` to handle images."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "vscode": {
     "languageId": "shellscript"
    }
   },
   "outputs": [],
   "source": [
    "!pip install datasets Pillow"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "And we need to login into Hugging Face to upload the dataset. You should goto the [Hugging Face website](https://huggingface.co/settings/tokens) to get your API token."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "vscode": {
     "languageId": "shellscript"
    }
   },
   "outputs": [],
   "source": [
    "!huggingface-cli login --token hf_YOUR_HF_TOKEN # replace hf_YOUR_HF_TOKEN to your own Hugging Face token."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "vscode": {
     "languageId": "plaintext"
    }
   },
   "source": [
    "## Download Dataset\n",
    "\n",
    "We have uploaded the zip file of the dataset to [Hugging Face](https://huggingface.co/datasets/pufanyi/VQAv2_TOY/tree/main/source_data) for download. This dataset is a subset of the [VQAv2](https://visualqa.org/) dataset, with $20$ entries each from the `val`, `test`, and `test-dev` splits, for easier downloading."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "vscode": {
     "languageId": "shellscript"
    }
   },
   "outputs": [],
   "source": [
    "!wget https://huggingface.co/datasets/lmms-lab/VQAv2_TOY/resolve/main/source_data/sample_data.zip -P data\n",
    "!unzip data/sample_data.zip -d data"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can open `data/questions` to take a view of the dataset organization. We found that the toy-`VQAv2` dataset is organized as follows:\n",
    "\n",
    "```json\n",
    "{\n",
    "    \"info\": { /* some infomation */ },\n",
    "    \"task_type\": \"TASK_TYPE\", \"data_type\": \"mscoco\",\n",
    "    \"license\": { /* some license */ },\n",
    "    \"questions\": [\n",
    "        {\n",
    "            \"image_id\": 262144, // integer id of the image\n",
    "            \"question\": \"Is the ball flying towards the batter?\",\n",
    "            \"question_id\": 262144000\n",
    "        },\n",
    "        /* ... */\n",
    "    ]\n",
    "}\n",
    "```"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Define Dataset Features _(Optional<sup>*</sup>)_\n",
    "\n",
    "You can define the features of the dataset. For more details, please refer to the [official documentation](https://huggingface.co/docs/datasets/en/about_dataset_features).\n",
    "\n",
    "<sup>*</sup> _Note that if the dataset features are consistent and all entries in your dataset table are non-null **for all splits of data**, you can skip this step._"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import datasets\n",
    "\n",
    "features = datasets.Features(\n",
    "    {\n",
    "        \"question_id\": datasets.Value(\"int64\"),\n",
    "        \"question\": datasets.Value(\"string\"),\n",
    "        \"image_id\": datasets.Value(\"string\"),\n",
    "        \"image\": datasets.Image(),\n",
    "    }\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Define Data Generator\n",
    "\n",
    "We use [`datasets.Dataset.from_generator`](https://huggingface.co/docs/datasets/v2.20.0/en/package_reference/main_classes#datasets.Dataset.from_generator) to create the dataset.\n",
    "\n",
    "The generator function should `yield` dictionaries with the keys corresponding to the dataset features. This can save memory when loading large datasets.\n",
    "\n",
    "For the image data, we can convert the image to [`PIL.Image`](https://pillow.readthedocs.io/en/stable/reference/Image.html) object.\n",
    "\n",
    "Note that if some columns are missing in some splits of the dataset (for example, the `answer` column is usually missing in the `test` split), we need to set these columns to null to ensure that all splits have the same features."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import json\n",
    "from PIL import Image\n",
    "\n",
    "\n",
    "def generator(qa_file, image_folder, image_prefix):\n",
    "    with open(qa_file, \"r\") as f:\n",
    "        data = json.load(f)\n",
    "        qa = data[\"questions\"]\n",
    "\n",
    "    for q in qa:\n",
    "        image_id = q[\"image_id\"]\n",
    "        image_path = os.path.join(image_folder, f\"{image_prefix}_{image_id:012}.jpg\")\n",
    "        q[\"image\"] = Image.open(image_path)\n",
    "        yield q"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Generate Dataset\n",
    "\n",
    "We generate the dataset using the generator function.\n",
    "\n",
    "Note that if you skip the step of defining dataset features, there is no need to pass the `features` argument. The dataset infer the features from the dataset automatically."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "NUM_PROC = 32  # number of processes to use for multiprocessing, set to 1 for no multiprocessing\n",
    "\n",
    "data_val = datasets.Dataset.from_generator(\n",
    "    generator,\n",
    "    gen_kwargs={\n",
    "        \"qa_file\": \"data/questions/vqav2_toy_questions_val2014.json\",\n",
    "        \"image_folder\": \"data/images\",\n",
    "        \"image_prefix\": \"COCO_val2014\",\n",
    "    },\n",
    "    # For this dataset, there is no need to specify the features, as all cells are non-null and all splits have the same schema\n",
    "    # features=features,\n",
    "    num_proc=NUM_PROC,\n",
    ")\n",
    "\n",
    "data_test = datasets.Dataset.from_generator(\n",
    "    generator,\n",
    "    gen_kwargs={\n",
    "        \"qa_file\": \"data/questions/vqav2_toy_questions_test2015.json\",\n",
    "        \"image_folder\": \"data/images\",\n",
    "        \"image_prefix\": \"COCO_test2015\",\n",
    "    },\n",
    "    # features=features,\n",
    "    num_proc=NUM_PROC,\n",
    ")\n",
    "\n",
    "data_test_dev = datasets.Dataset.from_generator(\n",
    "    generator,\n",
    "    gen_kwargs={\n",
    "        \"qa_file\": \"data/questions/vqav2_toy_questions_test-dev2015.json\",\n",
    "        \"image_folder\": \"data/images\",\n",
    "        \"image_prefix\": \"COCO_test2015\",\n",
    "    },\n",
    "    # features=features,\n",
    "    num_proc=NUM_PROC,\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Dataset Upload\n",
    "\n",
    "Finally, we group the dataset with different splits and upload it to the Huggingface dataset hub."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "data = datasets.DatasetDict({\"val\": data_val, \"test\": data_test, \"test_dev\": data_test_dev})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "data.push_to_hub(\"lmms-lab/VQAv2_TOY\")  # replace lmms-lab to your username"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now, you can check the dataset on the [Hugging Face dataset hub](https://huggingface.co/datasets/lmms-lab/VQAv2_TOY)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "lmms-eval",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.11.9"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}