fireviewer/fire-smoke-detection-corpus-v1

FireViewer Fire/Smoke Detection Corpus v1

Quick card

Field	Reference
Use	Inspect the historical rights-filtered visible smoke/flame detection corpus.
Expected input	Image rows and source-specific object labels following the published schema.
Example	Stream one pinned split; preserve each row's source and annotation provenance.
Limits	Historical release with explicit source removals; public membership does not grant new source rights.
Contract	Schema · Pinned loading

Choose a variant · Resource navigation · Update log

Status

Versioned detector-training corpus. Current V1 repository contents: 87,568 rows, split 53,952 train / 12,787 validation / 20,829 test.

The corpus stores source-specific provenance, hashes, grouping/de-duplication information, validation status and annotation metadata. It preserves the V1 training reference and provenance used by earlier strict FireViewer detector releases; each model receipt identifies the snapshot actually used.

Rights

There is no single licence covering all source images. Keep repository-level license: other and preserve the source licence/attribution recorded for every sample. FireViewer publication must not overwrite source obligations.

Use boundary

This corpus is a training/research resource. It is not an incident report, emergency feed or evidence that every image is suitable for every deployment domain.

Sources, labels and exclusions

The original corpus audit records the historical 102,257-row release. The current clean shard counts and file hashes are in DATA_FILES_MANIFEST.json, and the exact removal is in CLEANUP_RECEIPT.json.

Source	Records	Recorded source licence
fasdd	57,362	CC-BY-SA-4.0
pyro-sdis	30,206	Apache-2.0

These are the obligations recorded in the source manifests, not a new legal assessment. Read source rights matrix before redistribution.

The release preserves admitted manifest splits and excludes quarantine, review and historical-overlap rows. The 17 September duplicate audit found no duplicate sample IDs or recorded image SHA-256 values in the four complete manifests. It did not redownload/re-hash image payloads or establish absence of every near-duplicate. Source annotations and automatic validation must not be relabelled as a fresh human review.

Schema

Parquet rows include image, sample_id, source_name, source IDs, dimensions, sha256, split, split_group, licence, negative, validation status and validation-run identity. Structured annotations and source records are stored in JSON string fields such as annotations_json and original_record_json. Parse those strings before use and preserve class mapping and geometry conventions from the source record; do not infer an undocumented box coordinate convention.

Loading a pinned split

The repository contains published Parquet shards under data/train/, data/validation/ and data/test/. Audit and rights files are metadata, not training examples.

from datasets import load_dataset

ds = load_dataset(
    "fireviewer/fire-smoke-detection-corpus-v1",
    revision="32fe9b9205822f104da296b9d5053a65f59cf6f4",
    data_files={"train": "data/train/*.parquet"},
    split="train",
    streaming=True,
)

Streaming avoids materializing the complete corpus locally. Keep grouping and splits intact; the test split must not become a model-selection set. Data-file hashes, source evidence and publication manifest provide the retained provenance. Structural or automatic checks do not establish field accuracy or scientifically qualified pointing labels.

September 2026 source removal

The 26 September cleanup receipt records removal of Alarmod (8,403 rows) and Boréal (6,286 rows) from every Parquet split, reducing the repository from 102,257 to 87,568 rows. Their source manifests and provenance files were removed from the current tree. Historical audits and prior commits retain their original counts and do not describe the current sample set.

For the separately packaged v2.5 detection release, see Visible Smoke and Flame Detection v2.5. Its labels, splits and provenance are defined by its own release rather than by this V1 card.

Documentation and revision

Card updated in English on 5 October 2026, from the public repository at 32fe9b9205822f104da296b9d5053a65f59cf6f4 and its retained reports. This update checks documentation, repository metadata and small evidence files; it does not rerun training, inference, dataset payload verification or device qualification. Historical receipts keep their original dates, revisions and scope. Earlier README hashes in artifact manifests refer to those earlier releases.

The association FIRE-VIEWER provides administrative and financial stewardship for resources under its control. Technical governance is maintainer-led. Upstream, source-specific and pre-association rights remain separate; repository placement does not transfer ownership or grant additional rights.

Project documentation · Institutional website. Contact: contact@fire-viewer.fr.

Likes are always welcome! ❤️ If these resources help you, a little heart makes our day and helps others discover the work. Thank you!