Datasetpublic

grantmwilkinson/epri-transmission-ttpla

EPRI Transmission TTPLA Imagery Aerial imagery of transmission towers and power lines with polygon (instance-segmentation) annotations, TTPLA-style. 1,242 images, 3,840 × 2,160 px. All annotations are polygons. 70 / 15 / 15 train / val / test split (seed 42). Classes id class geometry supercategory is_line 1 cable polygon line false 2 tower_lattice polygon structure false 3 tower_tucohy polygon structure false 4 tower_wooden polygon structure… See the full description on the dataset page: https://huggingface.co/datasets/grantmwilkinson/epri-transmission-ttpla.

sourceHugging Faceotherupdated 3mo agoView on Hugging Face

0likes334downloads

Dataset Card

EPRI Transmission TTPLA Imagery

Aerial imagery of transmission towers and power lines with polygon (instance-segmentation) annotations, TTPLA-style.

—1,242 images, 3,840 × 2,160 px.
—All annotations are polygons.
—70 / 15 / 15 train / val / test split (seed 42).

Classes

id	class	geometry	supercategory	is_line
1	cable	polygon	line	false
2	tower_lattice	polygon	structure	false
3	tower_tucohy	polygon	structure	false
4	tower_wooden	polygon	structure	false
5	void	polygon	ignore	false

The mis-cased source label Cable was normalized to cable. void marks ignore / occlusion regions.

Schema

Each row: image, image_id, file_name, width, height, and annotations (columnar lists): id, category_id, category_name, supercategory, segmentation (flat [x0,y0,x1,y1,...]), bbox ([x,y,w,h]), area, iscrowd, is_line, subclass_id, subclass_name (TTPLA has no occlusion tag → 0 / none).

Merging with the H-frame dataset

Shares its schema with `grantmwilkinson/epri-transmission-hframe`. Filter supercategory == "line" to select every wire across both datasets. Here cables are filled polygons (is_line = false) — rasterize with cv2.fillPoly; the H-frame wires are polyline centerlines (is_line = true, use cv2.polylines(..., thickness=k)).

Please consult the original TTPLA dataset terms before redistribution or commercial use.