metadata

license: cc-by-4.0
dataset_info:
  features:
    - name: Time [s]
      dtype: float32
    - name: id
      dtype: int32
    - name: Type
      dtype: string
    - name: x_img [px]
      dtype: int32
    - name: y_img [px]
      dtype: int32
    - name: Angle_img [rad]
      dtype: float32
    - name: Frame
      dtype: string
    - name: Image
      dtype: image
    - name: Mask
      dtype: image
  splits:
    - name: train
      num_bytes: 1768280
      num_examples: 273
  download_size: 1792775
  dataset_size: 1768280
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
task_categories:
  - image-segmentation
tags:
  - aerial
  - vehicles
  - drones

ORD for the Sciences Hackathon - Vehicles Detection

This project is an example of a hackathon project. The quality of the data produced has not been evaluated. Its goal is to provide an example on how a dataset can be update to Hugginface.

This is an example of a hackathon project presented to ORD for the sciences hackathon using the openly available pNeuma vision dataset.

Description

The goal of this project is to create a training dataset derived from the publicly available pNeuma Vision dataset, which contains drone footage and coordinates of vehicles. By leveraging machine learning techniques, specifically the "Segment Anything" model by Meta, we will accurately segment and mask the pixels corresponding to each vehicle within the footage. The resulting dataset, stored in the efficient Parquet format, will be shared on Hugging Face as a new, open-access resource for the research community. Additionally, we will document our methodology in a detailed Jupyter notebook, which will be hosted in a public GitHub repository. Our work will be registered as a derived contribution in the pNeuma RDI Hub prototype, ensuring proper attribution and fostering further research and development.

Datasets created: