Emulated-Inc/contrailseg

Contrail Segmentation in Infrared Satellite Imagery

Find contrails, thin line-shaped ice clouds, in 256 by 256 nine-channel infrared brightness-temperature images. Each record is one labelled frame stored as two uint8 planes (hi and lo) of a 4 mK quantised brightness-temperature grid, with channels named by wavelength. Training records carry run-length-encoded consensus masks; the held-out records carry images only.

Contents

The data tree lives under data/, exactly as the benchmark environment presents it at /app/contrailseg. Its own description is at data/README.md.

• 22389 files, 16704970306 bytes (15.6 GiB)
• structure hash (sha256 over sorted path\tsize lines): d9fe61308b60c4f5e475776924fb89b5663526c3463591bf80662a19299898a5
• data.manifest.tsv at the repo root lists every file as sha256 size path

Use

from huggingface_hub import snapshot_download

snapshot_download("Emulated-Inc/contrailseg", repo_type="dataset", revision="REVISION",
                  local_dir="./contrailseg", allow_patterns="data/*")

Pin revision to a commit sha rather than a branch if you need reproducibility.

Provenance and licence

Derived from the OpenContrails dataset published by Google Research under CC BY 4.0 (GCS bucket goes_contrails_dataset, snapshot 20230419), the data of the Kaggle competition Google Research - Identify Contrails to Reduce Global Warming (https://www.kaggle.com/competitions/google-research-identify-contrails-reduce-global-warming). Please cite Ng et al., "OpenContrails: Benchmarking Contrail Detection on GOES-16 ABI", arXiv:2304.02122 (2023). The underlying GOES-16 ABI imagery is unrestricted. The held-out split used by the benchmark this dataset serves is carved from the competition's labelled data; its labels are not included here.