{
"fetched_at": "2026-10-19T21:28:23Z",
"source_url": "https://www.kaggle.com/competitions/electric-transmission-imagery/data",
"pages": [
{
"name": "data-description",
"content": "#Summary of data\nThe Electric Transmission Imagery data contains drone-collected images collected during a electric utility inspection. Here is a summary of each dataset and a list of key variables with descriptions.\n\nclassid --> {'crossarm':0,'h_structure':1,'insulator':2,'pole':3,'x_brace':4, 'conductor': 'shield wire':6}\n\nxmin --> minimum value of x coordinate of bounding box\n\nymin --> minimum value of y coordinate of bounding box\n\nwidth--> width of bounding box\n\nheight--> height of bounding box\n\nsubclassificationid --> {'none':0,'both':1,'obscured':2,'truncated':3}\n\n#File descriptions\n`README_train.txt` provides a summary of the GTstring format. \n`train.csv` contains a list of 8038 image file names (ImageID) and bounding box annotations (GTstring) for each image.\n\n`test.csv` contains a list of 3445 image file names (ImageID) to be used for model testing.\n\n`README_submission.txt` provides a summary of the submission syntax.
\n`submission_sample.csv` provides an example format for the submission syntax.\n\n*This competition does not accept any predictions on conductor or shield wire classes (polylines). However, EPRI believes there is value in sharing the data and may run a competition on this task in the future.*
\n`README_train_polyline.txt` provides a summary of the GTstring format.
\n`train_polyline.csv` contains a list of 6683 unique file names (ImageID) and polyline annotations (GTstring) for each image\n\n#Imagery\nThe Transmission H-frame Dataset 1.0 dataset contains 12,158 drone collected images of one transmission structure type, an \"H-frame\" structure type. 11,483 have been labeled and split 8,038 for public training and 3,445 for withheld testing.\nThese images can be obtained by completing the form below.\nhttps://forms.office.com/r/2Qy0n4NnVA\n\n",
"isPublished": false,
"hasIsPublished": false,
"postTitle": "",
"hasPostTitle": false,
"mimeType": "",
"hasMimeType": false
},
{
"name": "Description",
"content": "The Electric Power Research Institute (EPRI) is building a community to advance utility infrastructure inspection. To jumpstart collaboration EPRI is sharing labeled datasets. These data may support immediate and future research. Advances in automated inspection workflows can benefit society. Future tools may revolutionize traditional workflows and have safety, quality, and efficiency benefits to utilities and consumers. More situational awareness of eclectic utility infrastructure can reduce the likelihood of in-service equipment failures.\n\nVisually inspecting high voltage electrical infrastructure is complex. There are several components and areas of interest each with unique degradation modes. Automated inspection technologies may require large datasets to handle the real-world complexities. It is very likely these datasets do not exist in 2022. EPRI is addressing the data gap by creating and sharing datasets to support research and commercial innovation. \n\nThe **Transmission H-frame Dataset 1.0** dataset contains 12,158 drone collected images of one transmission structure type, an \"H-frame\" structure type. 11,483 have been labeled and split 8,038 for public training and 3,445 for withheld testing.\nThese images can be obtained by completing the form below.\nhttps://forms.office.com/r/7uuD1putv1 \n\n##Data Description\n\n\n\nThe **Transmission H-frame Dataset 1.0** dataset contains 12,158 drone collected images of one transmission structure type, an \"H-frame\" structure type. 11,483 have been labeled and split 8,038 for public training and 3,445 for withheld testing.
**These images can be obtained by completing the form below.**\nhttps://forms.office.com/r/2Qy0n4NnVA\n\nAfter the form is complete, an email is generated containing hyperlinks to the image datasets. The full resolution (5280x2970) imagery, as well as, lower resolution versions will be shared. The imagery has been labeled with seven classes. Five of these are bounding boxes, and two are polylines defined by points. Each class contains metadata denoting obscuration, truncation, or both. \n\n\nSome image pixels have been obscured to limit personally identifiable information. Features include faces, licenses plates, company logos, and other objects based on labeler discretion. EPRI requests that if you identify an image that you think contains sensitive information, please email AI@epri.com with subject \"Sensitive Image, H_Structure_Vx, Hframe(xxxxx).jpg\". Should any edits be made, EPRI will update the dataset accordingly. \n\n\nAdditionally, the GPS information has been translated in an attempt to anonymize the location. Relative position information between each image remains. The code used for the GPS translation can be found here, https://github.com/pkulkarni-epri/DistributionTaxonomy/blob/main/manipulate_exif.py\n\n\n## Acknowledgements\n\nThe electric utilities that support EPRI's research project 1-114205, AI for Transmission Images.",
"isPublished": false,
"hasIsPublished": false,
"postTitle": "",
"hasPostTitle": false,
"mimeType": "",
"hasMimeType": false
},
{
"name": "rules",
"content": "There is no prize within this community prediction competition. EPRI's intent is to share electric utility data to support public, private, and academic research and innovation. The imagery and data shared within this project is shared under Creative Commons Attribution, Share-a-Like licenses. \n\nAttribution\nDataset: Transmission H-frame dataset 1.0\nCreator: EPRI, P. Kulkarni, D. Lewis\nLicense: CC BY-SA 4.0\n\nAs such, the \"rules\" are fairly simple. \n\n - Don't cheat!\n - Apply yourself!\n - Have fun!",
"isPublished": false,
"hasIsPublished": false,
"postTitle": "",
"hasPostTitle": false,
"mimeType": "",
"hasMimeType": false
},
{
"name": "Labeling Guidance – Polywines",
"content": "\n\n\n\n",
"isPublished": false,
"hasIsPublished": false,
"postTitle": "",
"hasPostTitle": false,
"mimeType": "",
"hasMimeType": false
},
{
"name": "Labeling Guidance – Bounding Boxes",
"content": "\n\n\n\n\n\n\n\n\n\n\n\n",
"isPublished": false,
"hasIsPublished": false,
"postTitle": "",
"hasPostTitle": false,
"mimeType": "",
"hasMimeType": false
},
{
"name": "Labeling Guidance Scenarios And Relationships",
"content": "\n\n\n\n\n\n\n\n",
"isPublished": false,
"hasIsPublished": false,
"postTitle": "",
"hasPostTitle": false,
"mimeType": "",
"hasMimeType": false
},
{
"name": "Acknowledgement",
"content": "\n",
"isPublished": false,
"hasIsPublished": false,
"postTitle": "",
"hasPostTitle": false,
"mimeType": "",
"hasMimeType": false
},
{
"name": "Other Imagery",
"content": "https://www.kaggle.com/datasets/dexterlewis/epri-distribution-inspection-imagery\n",
"isPublished": false,
"hasIsPublished": false,
"postTitle": "",
"hasPostTitle": false,
"mimeType": "",
"hasMimeType": false
},
{
"name": "Evaluation",
"content": "This competition will focus on object detection of the five bounding box classes only. A custom multiclass/column root mean squared metric must be used to measure intersection over union pixel error within an automated Kaggle workflow. \n\n\n\n\n",
"isPublished": false,
"hasIsPublished": false,
"postTitle": "",
"hasPostTitle": false,
"mimeType": "",
"hasMimeType": false
}
]
}