{"id":2952878,"date":"2026-01-26T03:54:48","date_gmt":"2026-01-26T11:54:48","guid":{"rendered":"https:\/\/www.esri.com\/arcgis-blog\/?post_type=blog&p=2952878"},"modified":"2026-02-25T11:03:50","modified_gmt":"2026-02-25T19:03:50","slug":"use-external-object-detection-training-data-within-arcgis","status":"publish","type":"blog","link":"https:\/\/www.esri.com\/arcgis-blog\/products\/arcgis-pro\/geoai\/use-external-object-detection-training-data-within-arcgis","title":{"rendered":"Use external object detection training data within ArcGIS"},"author":254042,"featured_media":0,"parent":0,"menu_order":0,"comment_status":"open","ping_status":"closed","template":"","format":"standard","meta":{"_acf_changed":false,"_searchwp_excluded":""},"categories":[770712,22931],"tags":[780852,186132,515302,780851,780853],"industry":[],"product":[36571,36581,36551,36561,42661,36601],"class_list":["post-2952878","blog","type-blog","status-publish","format-standard","hentry","category-geoai","category-imagery","tag-coco","tag-deep-learning","tag-object-detection","tag-pascal-voc","tag-yolo","product-arcgis-enterprise","product-arcgis-living-atlas","product-arcgis-online","product-arcgis-pro","product-arcgis-solutions","product-developers"],"acf":{"short_description":"Learn how to convert YOLO, COCO, and AzureML data to PASCAL VOC for seamless deep learning in ArcGIS using our Python scripts.","flexible_content":[{"acf_fc_layout":"content","content":"

We designed ArcGIS deep learning tools to work seamlessly with training data in the PASCAL VOC<\/strong> (P<\/strong>attern A<\/strong>nalysis, S<\/strong>tatistical Modeling and C<\/strong>omputational L<\/strong>earning, V<\/strong>isual O<\/strong>bject C<\/strong>lasses) format. While we recommend creating and managing your training data directly within ArcGIS to ensure full compatibility and reliability, we understand that you may need to work with datasets originally labeled in other formats.<\/p>\n

To help you work with your existing datasets and streamline your deep learning workflows, we\u2019ve developed a suite of Python scripts that convert metadata from several common formats into ArcGIS-ready PASCAL VOC formats.<\/p>\n

Why PASCAL VOC?<\/h2>\n

The PASCAL VOC format is a computer vision standard that keeps things simple. Its XML files<\/b> specify bounding boxes using absolute pixel coordinates ( <xmin><\/code>, <ymin><\/code>, <xmax><\/code>, and <ymax><\/code> ) rather than percentages or ratios.<\/p>\n

These absolute corner points differentiate PASCAL VOC from other formats. While external data varies in file type (JSON, TXT, etc.) and structure, the biggest difference lies in how they define the bounding box. Many formats record normalized values (0.0 to 1.0) or relative dimensions (center point, width, height) instead of pixels. Because ArcGIS tools require the PASCAL VOC standard, you must convert these external datasets to absolute pixel coordinates to use them successfully.<\/p>\n

 <\/p>\n"},{"acf_fc_layout":"image","image":{"ID":2943899,"id":2943899,"title":"Example of PASCAL VOC","filename":"pascal_voc-e1767125352602.jpg","filesize":25920,"url":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2025\/10\/pascal_voc-e1767125352602.jpg","link":"https:\/\/www.esri.com\/arcgis-blog\/products\/arcgis-pro\/geoai\/converting-object-detection-labels-to-pascal-voc-to-use-in-arcgis\/pascal_voc","alt":"Example of PASCAL VOC","author":"254042","description":"Example of PASCAL VOC","caption":"Example of PASCAL VOC","name":"pascal_voc","status":"inherit","uploaded_to":2943891,"date":"2025-10-19 00:55:40","modified":"2026-01-14 18:51:48","menu_order":0,"mime_type":"image\/jpeg","type":"image","subtype":"jpeg","icon":"https:\/\/www.esri.com\/arcgis-blog\/wp-includes\/images\/media\/default.png","width":600,"height":430,"sizes":{"thumbnail":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2025\/10\/pascal_voc-213x200.jpg","thumbnail-width":213,"thumbnail-height":200,"medium":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2025\/10\/pascal_voc-e1767125352602.jpg","medium-width":364,"medium-height":261,"medium_large":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2025\/10\/pascal_voc-e1767125352602.jpg","medium_large-width":600,"medium_large-height":430,"large":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2025\/10\/pascal_voc-e1767125352602.jpg","large-width":600,"large-height":430,"1536x1536":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2025\/10\/pascal_voc-e1767125352602.jpg","1536x1536-width":600,"1536x1536-height":430,"2048x2048":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2025\/10\/pascal_voc-e1767125352602.jpg","2048x2048-width":600,"2048x2048-height":430,"card_image":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2025\/10\/pascal_voc-649x465.jpg","card_image-width":649,"card_image-height":465,"wide_image":"https:\/\/www.esri.com\/arcgis-blog\/app\/uploads\/2025\/10\/pascal_voc-e1767125352602.jpg","wide_image-width":600,"wide_image-height":430}},"image_position":"center","orientation":"horizontal","hyperlink":""},{"acf_fc_layout":"content","content":"

\nClick here to view technical specifications for key external datasets<\/em><\/strong><\/summary>\n\n\n\n\n\n\n
\n

YOLO (You Only Look Once)<\/strong><\/h4>\n

[x_center, y_center, width, height]<\/code> where x_center<\/code> and y_center<\/code> represent the box center coordinates, and width<\/code> and height<\/code> specify the box size. All values are normalized as fractions of the image width and height.<\/p>\n

File Type:<\/strong> .txt<\/code><\/td>\n

\"yolo\"<\/td>\n<\/tr>\n
\n

COCO (Common Objects in Context)<\/strong><\/h4>\n

[x, y, width, height]<\/code> where x<\/code> and y<\/code> denote the top-left corner, and width<\/code> and height<\/code> specify the box size. All values are normalized as fractions of the image width and height.<\/p>\n

File Type:<\/strong> .json<\/code>(\u00a0a single file for the whole dataset)<\/td>\n

\"COCO\"<\/td>\n<\/tr>\n
\n

Azure Machine Learning<\/strong><\/h4>\n

[xmin, ymin, xmax, ymax]<\/code> where [xmin, ymin]<\/code>denotes the top-left corner and [xmax, ymax]<\/code>denotes the bottom-right corner. All values are normalized as fractions of the image width and height.
\nFile Type:<\/strong> .jsonl<\/code>(Single file with one JSON object per line)<\/td>\n

\"Azure<\/td>\n<\/tr>\n
\n

TFRecords<\/strong><\/h4>\n

[xmin, ymin, xmax, ymax]<\/code> where [xmin, ymin]<\/code>denotes the top-left corner and [xmax, ymax]<\/code>denotes the bottom-right corner. All values are normalized as fractions of the image width and height.<\/p>\n

File Type:<\/strong> .tfrecord<\/code>(a binary format)<\/td>\n

\"TFRecords\"<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/details>\n"},{"acf_fc_layout":"content","content":"

Convert to PASCAL VOC<\/h2>\n

External object detection training data must be converted to the PASCAL VOC absolute pixel standard for use in ArcGIS. Our Python scripts automate this transformation.<\/p>\n

Use the links below to download the relevant conversion script and documentation:<\/p>\n