-
Print
-
DarkLight
-
PDF
Download Data
-
Print
-
DarkLight
-
PDF
Export Annotations JSON
Each entity in our system has a JSON code that represents its data within our system.
This JSON code includes the entity's data related to our software and its metadata.
JSON is a minimal text-based data exchange format that is used primarily to transmit data between a server and web application. To understand more about JSON go here.
The following articles examplify and detail the JSON format for annotations, video annotations, and items:
Export annotations JSON for entire Dataset
In the project overview page, click the 3-dot action button for a specific dataset. Then select “Export Data” to download a ZIP file containing a separate JSON file per item in the dataset.
Export annotations JSON for filtered/queried/selected items
Exporting annotations JSON files for specific items from a dataset (either filtered, searched for, or manually selected) is done from the Dataset browser.
Click the download icon and select from the pop up dialog box
a. the scope of the export – whether you wish to export the entire dataset, selected items, or active search query
b. the content of the export – whether to include the annotations json files
Check “Include PNG per semantic label” to to add PNG files of semantic masks to the ZIP. It includes a PNG per mask (label) and a PNG for all masks (labels) combined.
Export COCO/YOLO/VOC annotations
In addition to the Dataloop format, annotations can be exported in standard formats: YOLO, COCO, and VOC.
Supports only Bounding Boxes without attributes.
Output should look like this (LabelNumber <x> <y> <width> <height>):
1 0.1930234375 0.1896117647058823 0.04320312499999999 0.19376470588235292
1 0.44302343750000006 0.18615294117647055 0.06342187500000006 0.23390588235294116
2 0.8295 0.2048470588235294 0.1158125000000001 0.32941176470588235
Supports only Bounding Boxes with attributes.
Output should look like this:
<annotation>
<folder></folder>
<filename>000000000785.jpg</filename>
<path>/000000000785.jpg</path>
<source>
<database>Unknown</database>
</source>
<size>
<width>640</width>
<height>425</height>
<depth>3</depth>
</size>
<segmented>0</segmented>
<object>
<name>aaa.abc</name>
<attributes>['all']</attributes>
<bndbox>
<xmin>178.53</xmin>
<ymin>353.53</ymin>
<xmax>485.0</xmax>
<ymax>379.40999999999997</ymax>
</bndbox>
</object> <object>
<name>aaa.abc</name>
<attributes>[]</attributes>
<bndbox>
<xmin>264.41674989846524</xmin>
<ymin>366.4664705882353</ymin>
<xmax>542.6488235294119</xmax>
<ymax>407.0564705882353</ymax>
</bndbox>
</object> <object>
<name>aaa</name>
<attributes>[]</attributes>
<bndbox>
<xmin>0.0</xmin>
<ymin>288.82529411764705</ymin>
<xmax>640.0</xmax>
<ymax>344.7052941176471</ymax>
</bndbox>
</object>
</annotation>
Is used for Bounding box, Polygons and Semantic annotations and does not support attributes. Click here for more information about the COCO data format.
Output should look like this:
{
"images": [
{
"file_name": "buffalo.jfif",
"id": 0,
"width": 266,
"height": 189
}
],
"info": {
"description": "test-filter"
},
"annotations": [
{
"bbox": [
87.61609907120743,
36.0945516235873,
16.408668730650163,
25.67136045563028
],
"segmentation": [
[
87.61609907120743,
41.66183461396495,
92.26006191950465,
36.0945516235873,
97.52321981424149,
38.568899619310706,
100.30959752321982,
42.2804216128958,
104.02476780185759,
58.98227058402874,
96.90402476780187,
61.76591207921758,
93.49845201238391,
60.52873808135587,
93.49845201238391,
54.961455090978234,
89.78328173374614,
51.24993309739312,
89.78328173374614,
47.2291176043426,
88.54489164086688,
45.9919436064809
]
],
"area": 289,
"iscrowd": 0,
"category_id": 2,
"image_id": 0,
"id": 0
}
],
"categories": [
{
"id": 1,
"name": "other"
},
{
"id": 2,
"name": "tree"
}
]
}
JSON Field | Content | Parent Key |
---|---|---|
images | an array of the images in the dataset | NA |
file_name | item filename | images |
id | the image unique id in this list; this id maps to the id field in the annotations array | images |
width | annotation width | images |
height | annotation height | images |
info | information about the dataset | NA |
description | dataset name | info |
annotations | the annotations used in each item | NA |
bbox | bounding box annotation coordinates | annotations |
segmentation | segmentation annotation coordinates | annotations |
area | the annotation area in pixels | annotations |
iscrowd | shows if the image contains a crowd of objects – a single object (iscrowd=0 in which case polygons are used) or a collection of objects (iscrowd=1 in which case RLE is used) | annotations |
category_id | the label identifier – maps to the id field of the categories array | annotations |
image_id | corresponds to the image id in the images array | annotations |
id | the annotation identifier – corresponds to the id in the images array | annotations |
categories | information about the labels available in the annotations | NA |
id | The label identifier – corresponds to the category_id field in an annotation object | categories |
name | the label name | categories |
- Got into the dataset browser of a specific dataset
- Click the “Applications” button and select the desired format - a ZIP file will be downloaded, containing JSON files for the items.
Single file item export
Annotations of a specific file can be exported in one of two ways:
- From the Dataset browser - select the file, and use the “Export data” button
- From the annotation studio
A. Open the file item in the annotation studio (from a specific task/assignment, or by double-clicking it in the dataset browser)
B. On the right-side panel, switch to the “ITEM” tab
C. Select the “Download Annotations” button and choose between "Export Annotations" and "Export mask".
+ Export Annotations - export annotation in Dataloop JSON format.
+ Export Mask - for items annotated with semantic segmentation, a mask file can be downloaded in PNG format.
File items download
The dataset browser enables downloading file items to your local file system.
This feature is available to users with project owner roles, and only when the data resides on the Dataloop storage. Items stored on external cloud storage (S3/Azure/GCS) cannot be downloaded.
- Quickly download a single item by right-clicking it and selecting download.
- For a more detailed download of a single item or multiple items, click the item/s you wish to download (to select multiple items hold Ctrl or Shift and click the items you wish to select) and click the download dataset icon
.
In the dialog box, check Selected items and the content you wish to download (e.g., Data item, Annotations JSON file, Include PNG per semantic label), and click Export.
The items will be downloaded to your local "Downloads" folder, without your dataset directory tree