Datasets:

fireviewer
/

firewarning-train-bundles-v1

Error code:   DatasetGenerationError
Exception:    CastError
Message:      Couldn't cast
integrity: struct<bundle_validations: list<item: null>, cross_source_exact_duplicates: int64, cross_source_spli (... 45 chars omitted)
  child 0, bundle_validations: list<item: null>
      child 0, item: null
  child 1, cross_source_exact_duplicates: int64
  child 2, cross_source_split_leakage: int64
  child 3, unique_image_sha256: int64
package_format: string
schema_version: int64
source_validation: list<item: struct<copied_files: int64, dataset_id: string, dataset_validation: struct<pair_inventory (... 651 chars omitted)
  child 0, item: struct<copied_files: int64, dataset_id: string, dataset_validation: struct<pair_inventory_verified:  (... 639 chars omitted)
      child 0, copied_files: int64
      child 1, dataset_id: string
      child 2, dataset_validation: struct<pair_inventory_verified: bool, raster_header_samples: struct<eo4: struct<count: int64, shape_ (... 430 chars omitted)
          child 0, pair_inventory_verified: bool
          child 1, raster_header_samples: struct<eo4: struct<count: int64, shape_max: list<item: int64>, shape_min: list<item: int64>>, hls: s (... 80 chars omitted)
              child 0, eo4: struct<count: int64, shape_max: list<item: int64>, shape_min: list<item: int64>>
                  child 0, count: int64
                  child 1, shape_max: list<item: int64>
                      child 0, item: int64
                  child 2, shape_min: list<item: int64>
                      child 0, item: int64
              child 1, hls: struct<cou
...
ing
  child 5, zip_size_bytes: int64
bundles: list<item: struct<train_id: string, artifact: struct<path: string, size_bytes: int64, sha256: string (... 493 chars omitted)
  child 0, item: struct<train_id: string, artifact: struct<path: string, size_bytes: int64, sha256: string, entry_cou (... 481 chars omitted)
      child 0, train_id: string
      child 1, artifact: struct<path: string, size_bytes: int64, sha256: string, entry_count: int64, crc_verified: bool, entr (... 24 chars omitted)
          child 0, path: string
          child 1, size_bytes: int64
          child 2, sha256: string
          child 3, entry_count: int64
          child 4, crc_verified: bool
          child 5, entry_sha256_verified: bool
      child 2, training_ready: bool
      child 3, sources: list<item: string>
          child 0, item: string
      child 4, evaluation_excluded: list<item: string>
          child 0, item: string
      child 5, dataset_ready: bool
      child 6, embedded_contract_status: string
      child 7, contract_revision: string
      child 8, blocking_reasons: list<item: string>
          child 0, item: string
      child 9, promotion_ready: bool
      child 10, promotion_blocking_reasons: list<item: string>
          child 0, item: string
      child 11, replacement_contract_revision: string
      child 12, do_not_execute_embedded_entrypoints: bool
      child 13, repair_note: string
      child 14, current_code_training_ready: bool
repository: string
repository_layout: string
to
{'schema_version': Value('int64'), 'repository': Value('string'), 'repository_layout': Value('string'), 'bundles': List({'train_id': Value('string'), 'artifact': {'path': Value('string'), 'size_bytes': Value('int64'), 'sha256': Value('string'), 'entry_count': Value('int64'), 'crc_verified': Value('bool'), 'entry_sha256_verified': Value('bool')}, 'training_ready': Value('bool'), 'sources': List(Value('string')), 'evaluation_excluded': List(Value('string')), 'dataset_ready': Value('bool'), 'embedded_contract_status': Value('string'), 'contract_revision': Value('string'), 'blocking_reasons': List(Value('string')), 'promotion_ready': Value('bool'), 'promotion_blocking_reasons': List(Value('string')), 'replacement_contract_revision': Value('string'), 'do_not_execute_embedded_entrypoints': Value('bool'), 'repair_note': Value('string'), 'current_code_training_ready': Value('bool')})}
because column names don't match
Traceback:    Traceback (most recent call last):
                File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1816, in _prepare_split_single
                  for key, table in generator:
                                    ^^^^^^^^^
                File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 613, in wrapped
                  for item in generator(*args, **kwargs):
                              ~~~~~~~~~^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 343, in _generate_tables
                  self._cast_table(pa_table, json_field_paths=json_field_paths),
                  ~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
                  pa_table = table_cast(pa_table, self.info.features.arrow_schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2369, in table_cast
                  return cast_table_to_schema(table, schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2297, in cast_table_to_schema
                  raise CastError(
                  ...<3 lines>...
                  )
              datasets.table.CastError: Couldn't cast
              integrity: struct<bundle_validations: list<item: null>, cross_source_exact_duplicates: int64, cross_source_spli (... 45 chars omitted)
                child 0, bundle_validations: list<item: null>
                    child 0, item: null
                child 1, cross_source_exact_duplicates: int64
                child 2, cross_source_split_leakage: int64
                child 3, unique_image_sha256: int64
              package_format: string
              schema_version: int64
              source_validation: list<item: struct<copied_files: int64, dataset_id: string, dataset_validation: struct<pair_inventory (... 651 chars omitted)
                child 0, item: struct<copied_files: int64, dataset_id: string, dataset_validation: struct<pair_inventory_verified:  (... 639 chars omitted)
                    child 0, copied_files: int64
                    child 1, dataset_id: string
                    child 2, dataset_validation: struct<pair_inventory_verified: bool, raster_header_samples: struct<eo4: struct<count: int64, shape_ (... 430 chars omitted)
                        child 0, pair_inventory_verified: bool
                        child 1, raster_header_samples: struct<eo4: struct<count: int64, shape_max: list<item: int64>, shape_min: list<item: int64>>, hls: s (... 80 chars omitted)
                            child 0, eo4: struct<count: int64, shape_max: list<item: int64>, shape_min: list<item: int64>>
                                child 0, count: int64
                                child 1, shape_max: list<item: int64>
                                    child 0, item: int64
                                child 2, shape_min: list<item: int64>
                                    child 0, item: int64
                            child 1, hls: struct<cou
              ...
              ing
                child 5, zip_size_bytes: int64
              bundles: list<item: struct<train_id: string, artifact: struct<path: string, size_bytes: int64, sha256: string (... 493 chars omitted)
                child 0, item: struct<train_id: string, artifact: struct<path: string, size_bytes: int64, sha256: string, entry_cou (... 481 chars omitted)
                    child 0, train_id: string
                    child 1, artifact: struct<path: string, size_bytes: int64, sha256: string, entry_count: int64, crc_verified: bool, entr (... 24 chars omitted)
                        child 0, path: string
                        child 1, size_bytes: int64
                        child 2, sha256: string
                        child 3, entry_count: int64
                        child 4, crc_verified: bool
                        child 5, entry_sha256_verified: bool
                    child 2, training_ready: bool
                    child 3, sources: list<item: string>
                        child 0, item: string
                    child 4, evaluation_excluded: list<item: string>
                        child 0, item: string
                    child 5, dataset_ready: bool
                    child 6, embedded_contract_status: string
                    child 7, contract_revision: string
                    child 8, blocking_reasons: list<item: string>
                        child 0, item: string
                    child 9, promotion_ready: bool
                    child 10, promotion_blocking_reasons: list<item: string>
                        child 0, item: string
                    child 11, replacement_contract_revision: string
                    child 12, do_not_execute_embedded_entrypoints: bool
                    child 13, repair_note: string
                    child 14, current_code_training_ready: bool
              repository: string
              repository_layout: string
              to
              {'schema_version': Value('int64'), 'repository': Value('string'), 'repository_layout': Value('string'), 'bundles': List({'train_id': Value('string'), 'artifact': {'path': Value('string'), 'size_bytes': Value('int64'), 'sha256': Value('string'), 'entry_count': Value('int64'), 'crc_verified': Value('bool'), 'entry_sha256_verified': Value('bool')}, 'training_ready': Value('bool'), 'sources': List(Value('string')), 'evaluation_excluded': List(Value('string')), 'dataset_ready': Value('bool'), 'embedded_contract_status': Value('string'), 'contract_revision': Value('string'), 'blocking_reasons': List(Value('string')), 'promotion_ready': Value('bool'), 'promotion_blocking_reasons': List(Value('string')), 'replacement_contract_revision': Value('string'), 'do_not_execute_embedded_entrypoints': Value('bool'), 'repair_note': Value('string'), 'current_code_training_ready': Value('bool')})}
              because column names don't match
              
              The above exception was the direct cause of the following exception:
              
              Traceback (most recent call last):
                File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 1369, in compute_config_parquet_and_info_response
                  parquet_operations, partial, estimated_dataset_info = stream_convert_to_parquet(
                                                                        ~~~~~~~~~~~~~~~~~~~~~~~~~^
                      builder, max_dataset_size_bytes=max_dataset_size_bytes
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                  )
                  ^
                File "/src/services/worker/src/worker/job_runners/config/parquet_and_info.py", line 948, in stream_convert_to_parquet
                  builder._prepare_split(split_generator=splits_generators[split], file_format="parquet")
                  ~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1683, in _prepare_split
                  for job_id, done, content in self._prepare_split_single(
                                               ~~~~~~~~~~~~~~~~~~~~~~~~~~^
                      gen_kwargs=gen_kwargs, job_id=job_id, **_prepare_split_args
                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                  ):
                  ^
                File "/usr/local/lib/python3.14/site-packages/datasets/builder.py", line 1869, in _prepare_split_single
                  raise DatasetGenerationError("An error occurred while generating the dataset") from e
              datasets.exceptions.DatasetGenerationError: An error occurred while generating the dataset

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

schema_version int64	repository string	repository_layout string	bundles list
1	Charlbi/firewarning-train-bundles-v1	one_zip_per_train	[ { "train_id": "media-filter-fire-smoke-v1", "artifact": { "path": "media-filter-fire-smoke-v1.zip", "size_bytes": 32772718037, "sha256": "ed6ee4cf71a2538cfce7b95e6e548528d94b4cea095c8d03ccfc4c0e5ba4c8ae", "entry_count": 173064, "crc_verified": true, "entry_sha256_verified...

FireViewer Training Bundles v1

Les identifiants FireWarning présents dans le slug, les noms d’archives, les manifestes, les stages et les empreintes sont conservés pour la compatibilité et la traçabilité. Ils ne désignent pas un projet actif distinct.

Fifteen self-contained, versioned ZIP64 packages prepared for the FireViewer training pipeline. Every training objective has exactly one ZIP. Each ZIP contains its source payloads, manifests, attribution and license metadata, entry-point contract, and a complete SHA-256 inventory under a single root directory.

Contrats à ne pas exécuter tels quels. Les ZIP publics fire-pointing-lora-v1.zip et cross-view-localization-v1.zip conservent leur ancien TRAIN_BUNDLE.json. Les données et empreintes restent valides, mais les entrypoints doivent être remplacés par la révision mvp-a40-v2 décrite dans les manifestes versionnés du dépôt. Aucune reconstruction ni republication de ces deux ZIP n'est effectuée dans cette passe.

Training objective	ZIP	Size (bytes)	SHA-256	State
Fire/smoke media filter	`media-filter-fire-smoke-v1.zip`	32,772,718,037	`ed6ee4cf71a2538cfce7b95e6e548528d94b4cea095c8d03ccfc4c0e5ba4c8ae`	Training-ready
Burned-area segmentation	`burned-area-segmentation-v1.zip`	27,926,097,197	`85c2f17248528ebbd5aa8395e72435ba5a12626bb5a53f5730109b11ea5dde36`	Dataset-ready; training blocked until independent geographic test
Cross-view localization	`cross-view-localization-v1.zip`	15,895,751,194	`5f5e14083da209bd978117e2c7470f8f63decdb4ba3223914da7e7164aa64f5f`	Data valid; DINOv2 entrypoint obsolete, rebuild required
Fire pointing LoRA	`fire-pointing-lora-v1.zip`	32,589,764,131	`a43cda497078b89960c300fce26280171441cf6111c208f96adfcfdecdc9762b`	Data valid; Qwen entrypoints obsolete, rebuild required
Cross-view registration	`cross-view-registration-v1.zip`	2,299,599,606	`a7e7d9205c34f96cf1843a7b7a9455e81a704c2926b9ba648bb00ecbddbc2f1e`	Dataset-ready; RoMa quality gate, double-validated test, and fine-tuning contract pending
Fire/smoke/normal media triage	`media-triage-fire-smoke-v1.zip`	31,688,094,244	`0ce5a30979007468efa8dd2f2ec5b20547943e080a6ed36318297023456ece34`	Dataset-ready; classifier trainer and independent critical test pending
Orchestrator gates SFT	`orchestrator-gates-sft-v1.zip`	17,444	`3ca4813cae6f04b806fce16be7a08a6854409dcd96df166d62a346e6d9a809ff`	Dataset-ready; SFT trainer and independent human validation pending
Fire progression/front inference	`fire-progression-front-inference-v1.zip`	6,493,320	`0bb682ee726d3ec62054e5c44b7dc84e0a0658e5e8ca5a1286662056020b7f11`	Dataset-ready; observed active-front labels and independent geographic test pending
Daily wildfire fact synthesis	`daily-wildfire-fact-synthesis-v1.zip`	222,162	`9e4c7dde666b964a3f38ae64c50383cd7faff7fbdab9d32a42b6ccca398fe332`	Dataset-ready; raw source documents and France validation pending
Structured wildfire situation/resources	`wildfire-situation-resources-structured-v1.zip`	8,696,884	`fc8bd5f419183b1ba603aab2117be842b75ad9acd46f9ba10eb2fff73e6ceb02`	Dataset-ready; France validation and stable incident identifiers pending
Engaged-assets object detection	`engaged-assets-object-detection-v1.zip`	1,193,478,095	`9a8e746f77ed88b7a8b3fa29ea13448f12ee9e8cdc6841ddfd8175950c079178`	Dataset-ready; fire-engine and operational-role labels pending
Camera depth and pose prior	`camera-depth-pose-prior-v1.zip`	39,593,395,922	`4f9574b2920126e7f34d62759ff0192c53062bc337e206309106beb0f670c280`	Dataset-ready; synthetic-only, real rural critical test pending
Outdoor metric depth	`outdoor-metric-depth-v1.zip`	132,759,914,158	`b9aa757c2e52007a3db40fc3c7f0878fb7dd391464832142e8d9b104cbb22bf2`	Training-ready as an auxiliary real outdoor depth prior
Wildfire smoke detection	`wildfire-smoke-detection-v1.zip`	13,518,198,597	`27abdbe3d3703d9c6fa67bfde136088845726313e148644eb0e948f9a6211e13`	Training-ready for smoke detection
Wildfire smoke segmentation	`wildfire-smoke-segmentation-v1.zip`	464,864,048	`23134190da8ef71b157764453f3d5575a339fe469878934c70e4972db33eee0e`	Dataset-ready; weak SAM masks and limited human masks

Total payload: 330,717,305,039 bytes.

Integrity

The accompanying *.validation.json and *.zip.sha256 files record the checks applied before publication. All fifteen archives passed:

full ZIP CRC verification;
SHA-256 verification of every entry;
a single training root per archive;
path-traversal and duplicate-entry rejection;
cross-source exact-duplicate and split-leakage checks.

The EO4Wildfires materialization contains exactly 31,730 scenes: 20,307 train, 5,077 validation, and 6,346 test.

The cross-view registration bundle contains 390 verified pairs split into 288 train, 57 validation, and 45 test rows across 14 isolated spatial groups. It includes 264 AerialExtreMatch examples and 126 French rural or mountain ODM examples. The private double-validation lot and the failed RoMa quality probe remain excluded.

The media-triage bundle derives image-level supervision only from verified source boxes and explicit negatives. It contains 155,044 images split into 107,449 train, 23,689 validation, and 23,906 test rows across 4,950 isolated groups. Touati and TaMduluza are not included: their published repository structure does not provide a reliable per-image class contract or a leakage-free video grouping.

The orchestrator bundle contains 118 deterministic decision examples generated by the current FireWarning stage contracts and gate engine: 110 stage-gate cases and 8 consensus cases. Its two synthetic daily workflows cover a nominal path and a contradiction that invokes the final judge and abstains when raw evidence is unavailable. It contains no operational Die or Fontainebleau fact and never authorizes automatic publication.

The FireSpread_MedEU bundle contains 103 event-isolated sequences and 316 usable cumulative burned-area targets. It does not claim observed active-front supervision. CrisisFACTS contributes 769 deduplicated facts over 26 event-days, while the structured IMSR bundle contains 15,982 incident-proxy sequences covering 88,208 incident-days; both remain out-of-domain for France until an independent French validation lot is approved.

The Open Images subset contains 4,398 licensed images split into 3,310 train, 435 validation, and 653 test samples, with no exact pixel leakage. It retains human-verified boxes for ambulances, helicopters, and fixed-wing aircraft. Open Images has no boxable fire-engine class and does not identify aircraft by firefighting role, so this bundle is supplemental rather than production-ready.

The TartanAir subset contains 34,742 aligned outdoor RGB, metric-depth, and camera-pose samples split by complete environment into 18,324 train, 5,838 validation, and 10,580 test rows. It is a synthetic geometry prior only: local NED poses are not geographic coordinates, and an independent real rural or mountain camera-pose test remains mandatory before operational use.

The DIODE outdoor bundle contains 17,330 real outdoor RGB images aligned with metric depth and validity masks, split by complete scene into 13,678 train, 105 validation, and 3,547 test rows. Its 15 scenes have no split-group leakage. DIODE's official downloadable test partition is not available through the public archive used here, so the FireWarning splits are deterministically assigned from complete train and validation scenes and recorded in the source manifest.

The Boreal detection subset contains 6,365 UAV images with 6,340 human smoke boxes and 256 documented empty-image negatives. Collection sites are isolated across 2,910 train, 2,217 validation, and 1,238 test rows. Fifty-two zero-byte labels outside the documented negative set are quarantined rather than silently converted into negative supervision. The segmentation subset contains 1,417 image-mask pairs: 40 human masks are strong labels and 1,377 SAM-generated masks remain weak supervision, recorded separately in every sample and split.

Licenses and attribution

This repository is a mixed-license collection. There is no single license covering every payload. Consult TRAIN_BUNDLE.json and the source metadata inside each ZIP before use or redistribution. Included source families declare licenses such as CC BY 4.0, CC BY-SA 4.0, CC0 1.0, Apache 2.0, MIT, and GPL 3.0, depending on the source.

Operational evaluation packages for Die and Fontainebleau are not included. Some openly licensed third-party training candidates can still document those locations; their original source URL, author, and license are retained in the corresponding manifest.

Publication policy

These packages are training inputs, not operational fire reports. They must not be used as evidence of a current incident, and generated locations or perimeters still require the FireWarning deterministic gates and human validation before publication.

Statut de publication FireViewer

Publication publique multi-licence. Quinze ZIP64 sont publiés, mais seulement trois bundles sont déclarés training_ready. Les douze autres conservent leurs blocages documentés. Les contrats fire-pointing-lora-v1 et cross-view-localization-v1 ne doivent pas être exécutés tels quels ; le contrat de remplacement est mvp-a40-v2.

Limites et sécurité

FireViewer est un projet de recherche et développement. Ce dataset n’est ni un service d’alerte, ni une source officielle, ni un outil de conduite des secours. Il ne doit pas être utilisé seul pour la sécurité des personnes, une décision d’intervention, une expertise d’assurance ou une preuve juridique.

Les observations, reconstructions, annotations et simulations doivent être interprétées avec leur provenance, leur confiance et leurs limites. Une interpolation rétrospective n’est ni une observation directe ni une prévision.

Droits

Chaque source garde sa licence propre, notamment CC BY 4.0, CC BY-SA 4.0, CC0, Apache-2.0, MIT, GPL-3.0 ou des conditions spécifiques. Aucune licence unique ne couvre automatiquement les quinze bundles.

Identité et contact

FireViewer est un projet distinct maintenu par Unicorn Who Dev. Les espaces fireviewer sur GitHub et Hugging Face sont ses espaces de publication canoniques.

Contact public, provenance, droits et demandes de retrait : unicornwhodev@gmail.com.

Downloads last month: 294

Total file size:

331 GB