Datasetpublic

Sliverwall/yolo26l_car_parts_queried

Multi-Component Image-to-Video Mapping (YOLO26L) Dataset Schema The data is stored in Apache Parquet format with snappy compression. Each row represents a continuous video clip that matches a specific query image. Column Type Description instance_id int64 Unique ID for the specific continuous sighting in the video. source_image_index int64 The index of the original query image from the reference aegean dataset. components_queried string The set of… See the full description on the dataset page: https://huggingface.co/datasets/Sliverwall/yolo26l_car_parts_queried.

sourceHugging Faceupdated 7mo agoView on Hugging Face

0likes3downloads

Dataset Card

Multi-Component Image-to-Video Mapping (YOLO26L)

Dataset Schema

The data is stored in Apache Parquet format with snappy compression. Each row represents a continuous video clip that matches a specific query image.

Column	Type	Description
`instance_id`	`int64`	Unique ID for the specific continuous sighting in the video.
`source_image_index`	`int64`	The index of the original query image from the reference aegean dataset.
`components_queried`	`string`	The set of car parts found in the query image.
`time_start`	`float64`	Start timestamp of the match in the video (seconds).
`time_end`	`float64`	End timestamp of the match in the video (seconds).
`mm_ss`	`string`	time range (e.g., `04:20 - 04:30`).
`url`	`string`	YouTube Embed URL with `?start=` and `?end=` parameters.

Retrieval Logic

The mapping was generated using the part-B notebook for assignment 2:

1.Reference images from the aegean dataset were processed by YOLO26L-seg at a 0.35 confidence threshold to extract component sets.
2.Performed a frame-level subset check against the video detection parquet file to find moments where the component set appeared
3.Continuous matches were grouped into "intervals" using a 5.0s gap threshold to account for occlusion