\n"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# Converting tags to classifications\n",
"\n",
"When working in the FiftyOne App, you can interactively [create, delete, and modify tags](https://voxel51.com/docs/fiftyone/user_guide/app.html#tags-and-tagging) right from the browser. When working with [Classification](https://voxel51.com/docs/fiftyone/user_guide/using_datasets.html#classification) labels, it makes sense to want to be able to annotate these labels the same as tags from the App.\n",
"\n",
"The primary distinction between [Classifications](https://voxel51.com/docs/fiftyone/user_guide/using_datasets.html#multilabel-classification) and [tags](https://voxel51.com/docs/fiftyone/user_guide/using_datasets.html#tags) is that Classifications can contain an arbitrary number of attributes associated with them, which is why it is generally recommended to use the [integrations with annotations tools like CVAT](https://voxel51.com/docs/fiftyone/integrations/cvat.html) to annotate these labels. Another benefit of using the [Classification](https://voxel51.com/docs/fiftyone/user_guide/using_datasets.html#classification) label type is that you can access the extensive [evaluation capabilities](https://voxel51.com/docs/fiftyone/user_guide/evaluation.html#classifications) of FiftyOne.\n",
"\n",
"However, if you are only populating the `label` of Classifications, then it can make sense to treat them like tags. This example shows how to convert the `label` of Classifications to tags, modify them in the App, and then convert tags back to Classifications. "
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Start by [installing FiftyOne](https://voxel51.com/docs/fiftyone/getting_started/install.html) if you haven't already:\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"!pip install fiftyone"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Then load an example dataset for this example."
]
},
{
"cell_type": "code",
"execution_count": 1,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Dataset already downloaded\n",
"Loading 'quickstart'\n",
" 100% |█████████████████| 200/200 [3.6s elapsed, 0s remaining, 51.3 samples/s] \n",
"Dataset 'quickstart' created\n"
]
}
],
"source": [
"import fiftyone as fo\n",
"import fiftyone.zoo as foz\n",
"from fiftyone import ViewField as F\n",
"\n",
"dataset = foz.load_zoo_dataset(\"quickstart\")"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Since this dataset doesn't contain classifications, let's start by converting the existing tags to classifications. The idea here is to efficient load all tags into memory using the [values()](https://voxel51.com/docs/fiftyone/user_guide/using_aggregations.html#values) aggregation, then converting the tags to classification labels and setting them to a field on the dataset."
]
},
{
"cell_type": "code",
"execution_count": 2,
"metadata": {},
"outputs": [],
"source": [
"# Convert all tags to Classifications\n",
"def tags_to_classifications(dataset, classifications_field):\n",
" tags = dataset.values(\"tags\")\n",
" classifications = []\n",
" for sample_tags in dataset.values(\"tags\"):\n",
" cls = []\n",
" if sample_tags:\n",
" for tag in sample_tags:\n",
" cls.append(fo.Classification(label=tag))\n",
" classifications.append(fo.Classifications(classifications=cls))\n",
" \n",
" dataset.set_values(classifications_field, classifications)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Note: You could also optionally add some logic here to choose a specific tag (like the first in the list) and create a Classification field rather than Classifications."
]
},
{
"cell_type": "code",
"execution_count": 3,
"metadata": {},
"outputs": [],
"source": [
"field_name = \"classifications\"\n",
"tags_to_classifications(dataset, field_name)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Now let's launch the App and modify some tags. For example, we can change the tag of the first two samples from `validation` to `test`."
]
},
{
"cell_type": "code",
"execution_count": 4,
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"\n",
"\n",
"\n",
"
![](\"https://user-images.githubusercontent.com/25985824/104791629-6e618700-5769-11eb-857f-d176b37d2496.png\")
\n",
" Try in Google Colab\n",
" \n",
" ![](\"https://user-images.githubusercontent.com/25985824/104791634-6efa1d80-5769-11eb-8a4c-71d6cb53ccf0.png\")
\n",
" Share via nbviewer\n",
" \n",
" ![](\"https://user-images.githubusercontent.com/25985824/104791633-6efa1d80-5769-11eb-8ee3-4b2123fe4b66.png\")
\n",
" View on GitHub\n",
" \n",
" ![](\"https://user-images.githubusercontent.com/25985824/104792428-60f9cc00-576c-11eb-95a4-5709d803023a.png\")
\n",
" Download notebook\n",
" \n",
" 
\n",
"