{"about":{"non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","site":"https://codewithpapers.app","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page","syntology":{"site":"https://syntology.ai","developers":"https://syntology.ai/developers","mcp":{"server":"https://syntology.ai/mcp","transport":"streamable-http","server_card":"https://syntology.ai/.well-known/mcp/server-card.json","auth":{"type":"trial token, no account","trial_token":"https://syntology.ai/api/oauth/trial/token","method":"POST","docs":"https://syntology.ai/developers"}},"have":"https://syntology.ai/api/graph/have?x=<method, arXiv id or title> (free, answers coverage only)","paper_base":"https://syntology.ai/paper/","atlas_base":"https://app.syntology.ai/?focus="},"machine_readable":[{"url":"https://codewithpapers.app/llms.txt","what":"the machine catalog: every machine-readable file, counted"},{"url":"https://codewithpapers.app/index/manifest.json","what":"paper-to-code index by arXiv id, with Syntology's counts"},{"url":"https://codewithpapers.app/search/manifest.json","what":"site search index (titles, authors) and its files"},{"url":"https://codewithpapers.app/download","what":"bulk files: Syntology's layer, described there"},{"url":"https://codewithpapers.app/build_manifest.json","what":"the build record: inputs, counts, exclusions, probes"}]},"url":"/dataset/ladi-v2","name":"LADI v2","full_name":"Low Altitude Disaster Imagery v2","description_markdown":"# LADI Overview\r\n\r\nThe Low Altitude Disaster Imagery (LADI) dataset was created to address the relative lack of annotated post-disaster aerial imagery in the computer vision community. Low altitude post-disaster aerial imagery from small planes and UAVs can provide high-resolution imagery to emergency management agencies to help them prioritize response efforts and perform damage assessments. In order to accelerate their workflow, computer vision can be used to automatically identify images that contain features of interest, including infrastructure such as buildings and roads, damage to such infrastructure, and hazards such as floods or debris. \r\n\r\nVersion 1 of the LADI dataset was funded as part of a NIST public safety innovation accelerator effort to create datasets for public safety. It collected over 40k images from the Civil Air Patrol (CAP), which were annotated by crowdsourced workers providing labels from a hierarchical set of classes. The dataset was used in a number of NIST TRECVID challenges ([2020](https://arxiv.org/abs/2104.13473), [2021](https://www.nist.gov/publications/evaluating-multiple-video-understanding-and-retrieval-tasks-trecvid-2021), and [2022](https://arxiv.org/abs/2306.13118)) and provided a broad basis for classification of aerial post-disaster scenes. However, the authors found that the quality of the labels were inconsistent, stemming in part from the untrained nature of the crowdsource workers, as well as the potentially subjective definition of certain labels, such as \"damage\".\r\n\r\nFor LADI v2, the authors used CAP volunteers who were trained in the FEMA damage assessment process, and we collected damage labels using the defined [FEMA Preliminary Damage Assessment scale](https://www.fema.gov/disaster/how-declared/preliminary-damage-assessments): unaffected, affected, minor, major, destroyed. These damage levels have specific criteria, helping reduce the subjectivity of identifying whether a structure is damaged. We also provide pretrained classifiers to aid in replication, and serve as a basis for fine-tuning and potential deployments.\r\n\r\n## Getting Started\r\nThe LADI v2 dataset is available on Hugging Face at [MITLL/LADI-v2-dataset](https://huggingface.co/datasets/MITLL/LADI-v2-dataset). This is the recommended method.\r\n### Default configuration\r\nThe default configuration uses the `v2a` labels with images resized to fit within 1800x1200. This is the recommended configuration for most use cases.\r\n\r\n```python\r\nfrom datasets import load_dataset\r\n\r\nds = load_dataset(\"MITLL/LADI-v2-dataset\")\r\n```\r\n\r\n### Advanced Usage\r\nIf you need access to the full resolution images, the `v2` label set, or the `v1` dataset, you should load from the script revision. This will use a custom dataset loader script, which will require you to set `revision=script` and `trust_remote_code=True`.\r\n\r\nThe available configurations for the script are: `v2`, `v2a`, `v2_resized`, `v2a_resized`, `v1_damage`, and `v1_infra`.\r\n\r\nThe first time you load the dataset, you should pass `download_ladi=True`, which will download a local copy of the relevant dataset version to your local system at `base_dir`.\r\n```python\r\nfrom datasets import load_dataset\r\n\r\nds = load_dataset(\"MITLL/LADI-v2-dataset\", \"v2a_resized\",\r\n                revision=\"script\",\r\n                streaming=True,\r\n                download_ladi=True,\r\n                base_dir='./ladi_dataset',\r\n                trust_remote_code=True)\r\n```\r\nThis only needs to be done once, and subsequent calls can omit the argument, and it will automatically use your local copy.\r\n```python\r\nds = load_dataset(\"MITLL/LADI-v2-dataset\", \"v2a_resized\",\r\n                revision=\"script\",\r\n                streaming=True,\r\n                base_dir='./ladi_dataset',\r\n                trust_remote_code=True)\r\n``` \r\n\r\nYou can also manually access the dataset at https://ladi.s3.amazonaws.com/index.html, or via AWS S3 at `s3://ladi`, and the provided dataset class in `training/LADI-v2-dataset`. We recommend using the [LADI_v2_resized](https://ladi.s3.amazonaws.com/ladi_v2_resized.tar.gz) version which resizes the images to 1800x1200, which should be large enough for most applications, but drastically reduces the overall dataset file size.\r\n\r\nThe LADI v1 and v2 dataset files are hosted as part of the [AWS Open Data](https://registry.opendata.aws/ladi/) program.\r\n\r\n### Dataset Details\r\n\r\nThe LADI-v2 dataset is a set of aerial disaster images captured and labeled by the Civil Air Patrol (CAP). The images are geotagged (in their EXIF metadata). Each image has been labeled in triplicate by CAP volunteers trained in the FEMA damage assessment process for multi-label classification; where volunteers disagreed about the presence of a class, a majority vote was taken. The classes are:\r\n\r\n- bridges_any\r\n- bridges_damage\r\n- buildings_affected\r\n- buildings_any\r\n- buildings_destroyed\r\n- buildings_major\r\n- buildings_minor\r\n- debris_any\r\n- flooding_any\r\n- flooding_structures\r\n- roads_any\r\n- roads_damage\r\n- trees_any\r\n- trees_damage\r\n- water_any\r\n\r\nThe v2 dataset consists of approximately 10k images, split into a train set of 8k images, a validation set of 1k images, and a test test of 1k images. The train and validation sets are drawn from the same distribution (CAP images from federally-declared disasters 2015-2022), whereas the test set is drawn from events in 2023, which has a different distribution of event types and locations. This is done to simulate the distribution shift as new events occur each year.\r\n\r\n#### Dataset v2a\r\nThe `v2a` variant of the dataset presents the same images with a modified set of labels, where the damage categories for buildings have been compressed into two classes of `buildings_affected_or_greater` and `buildings_minor_or_greater`. We find that this task is easier and of similar practical value for triage purposes. The `bridges_damage` label has also been removed due to the low number of positive examples in the dataset.\r\n\r\n- bridges_any\r\n- buildings_any\r\n- buildings_affected_or_greater\r\n- buildings_minor_or_greater\r\n- debris_any\r\n- flooding_any\r\n- flooding_structures\r\n- roads_any\r\n- roads_damage\r\n- trees_any\r\n- trees_damage\r\n- water_any\r\n\r\n\r\n#### Supported Tasks\r\nThe images are labeled for multi-label classification, as any number of the elements listed above may be present in a single image.\r\n\r\n#### Data Structure\r\nA single example in the v2a dataset looks like this:\r\n\r\n```\r\n{\r\n'image': <PIL.PngImagePlugin.PngImageFile image mode=RGB size=1800x1200 at ...>,\r\n'bridges_any': False, \r\n'buildings_any': False, \r\n'buildings_affected_or_greater': False, \r\n'buildings_minor_or_greater': False, \r\n'debris_any': False, \r\n'flooding_any': False, \r\n'flooding_structures': False, \r\n'roads_any': False, \r\n'roads_damage': False, \r\n'trees_any': True, \r\n'trees_damage': True, \r\n'water_any': True\r\n}\r\n```\r\n\r\n\r\n### Pretrained Classifiers \r\nWe provide a set of pretrained classifiers on the LADI v2 dataset for downstream finetuning and deployment purposes. See the associated model cards on Hugging Face for instructions on how to use.\r\n- [LADI-v2-classifier-small](https://huggingface.co/MITLL/LADI-v2-classifier-small) - Recommended for deployment, based on [google/bit-50](https://huggingface.co/google/bit-50), trained on entire LADI v2 dataset\r\n- [LADI-v2-classifier-large](https://huggingface.co/MITLL/LADI-v2-classifier-large) - Based on [microsoft/swinv2](https://huggingface.co/microsoft/swinv2-large-patch4-window12to16-192to256-22kto1k-ft), trained on entire LADI v2 dataset\r\n\r\nWe also provide \"reference\" versions of each model which are trained only on the `train` split of the LADI v2 dataset, to aid in reproducing the results presented in the paper.\r\n- [LADI-v2-classifier-small-reference](https://huggingface.co/MITLL/LADI-v2-classifier-small) \r\n- [LADI-v2-classifier-large-reference](https://huggingface.co/MITLL/LADI-v2-classifier-large-reference)\r\n\r\n## Citations\r\n\r\n### LADI v2\r\nIf you use the LADI v2 dataset or classifiers, please cite the following:\r\n\r\n```tex\r\n@misc{ladi_v2,\r\n      title={LADI v2: Multi-label Dataset and Classifiers for Low-Altitude Disaster Imagery}, \r\n      author={Samuel Scheele and Katherine Picchione and Jeffrey Liu},\r\n      year={2024},\r\n      eprint={2406.02780},\r\n      archivePrefix={arXiv},\r\n      primaryClass={cs.CV}\r\n}\r\n```\r\n\r\n## Distribution Information\r\n\r\nDISTRIBUTION STATEMENT A. Approved for public release. Distribution is unlimited.\r\n\r\nThis material is based upon work supported by the Department of the Air Force under Air Force Contract No. FA8702-15-D-0001. Any opinions, findings, conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the Department of the Air Force.\r\n\r\n© 2024 Massachusetts Institute of Technology.\r\n\r\nThe software/firmware is provided to you on an As-Is basis\r\n\r\nDelivered to the U.S. Government with Unlimited Rights, as defined in DFARS Part 252.227-7013 or 7014 (Feb 2014). Notwithstanding any copyright notice, U.S. Government rights in this work are defined by DFARS 252.227-7013 or DFARS 252.227-7014 as detailed above. Use of this work other than as specifically authorized by the U.S. Government may violate any copyrights that exist in this work.","description_withheld":null,"homepage":"https://github.com/ladi-dataset/ladi-overview","introduced_date":"2024-06-04","introduced_date_note":null,"introduced_by":{"paper":"/paper/ladi-v2-multi-label-dataset-and-classifiers","title":"LADI v2: Multi-label Dataset and Classifiers for Low-Altitude Disaster Imagery","first_author":"Samuel Scheele","url":null},"license":{"name":"cc-by-4.0, MIT","url":null},"modalities":[{"name":"Images","url":"/datasets/modality/images"}],"tasks":[{"name":"Multi-Label Image Classification","url":"/task/multi-label-image-classification","datasets_with_task":"/datasets/task/multi-label-image-classification"}],"languages":[],"variants":["LADI v2"],"data_loaders":[],"num_papers_in_archive":1,"source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28"},"benchmarks":[],"papers":{"order":"date (newest first), then slug","population":"every paper with a leaderboard row on this dataset's benchmarks (the benchmark-backed subset): the archive's own papers-using-this-dataset list was never published, so this is not that list; num_papers_in_archive is the archive's own count","page":1,"pages_in_order":1,"rows_per_page":100,"shown":0,"of":0,"counts":{"papers_with_a_benchmark_row":0,"with_a_code_link":0,"where_syntology_ran_a_sample":0,"not_listed_spam_title":0,"listed":0,"listed_where_code_ran":0,"where_syntology_ran_a_sample_split":{"with_a_run_with_no_instrument_failure":0,"every_run_a_failure_of_syntologys_instrument":0,"listed_with_a_run_with_no_instrument_failure":0,"listed_every_run_a_failure_of_syntologys_instrument":0,"filter":{"states":["a run with no instrument failure","any run, instrument failures included"],"default":"a run with no instrument failure","note":"on the 'only where code ran' pages the default hides, in the browser, the rows where every run was a failure of Syntology's instrument; the second state shows them again. Rows are hidden, never re-ordered; these twins list every row"}},"definition":"distinct papers with at least one leaderboard row on this dataset's benchmarks; 'where Syntology ran a sample' counts papers with at least one harvested sample that ran, which is not a correctness claim"},"items":[],"full_list":"papers_with_a_benchmark_row: every listed paper, in this order"},"papers_with_a_benchmark_row":[],"papers_where_code_ran":0,"papers_where_code_ran_first_page":null,"papers_where_code_ran_caption":"We ran code from the paper's repository; we did not run it on this dataset or check it against this dataset's benchmarks.","pages":[],"row_route":"each paper's syntology object carries sample_list (this site's paper page, #ran), syntology_url (syntology.ai's page for the paper, only when that route probed live) and mcp (the agent call that lists the paper's harvested samples; not a verdict on the paper)","developers":"https://syntology.ai/developers","syntology_totals":{"read_at":"2026-10-05T00:36:11+00:00","papers_with_samples":0,"samples_harvested":0,"samples_ran":0,"samples_constructed":0,"samples_ran_checked":0,"samples_ran_instrument_failed":0,"samples_unverified":0,"pointer_only_for_licence":0,"papers_with_no_sample_that_ran":0,"papers_with_a_sample_that_ran":0,"papers_with_a_run_with_no_instrument_failure":0,"papers_where_every_run_was_an_instrument_failure":0,"note":"the per-paper counts of every listed paper (all pages), summed; not a rate"},"key_notes":{"samples_ran_checked":"legacy name, kept unchanged so existing readers do not break: it counts the samples that ran with no instrument failure (honoured, violated, and ran with no contract checked); it does not mean a contract was checked, and the pages print it as 'K with no instrument failure', not 'K checked'","samples_constructed":"a sub-count of the samples that ran, never subtracted from them and never a failure: an executed sample whose run returned an instance of its own class (fixture_out_type equals the entry name): the run built an object and did not compute a result (Syntology's RAN record, counts.constructed)"},"since_archive_judged":{"papers":0,"papers_placed":0,"judged_by":{"model":"a model","models_as_recorded":[],"from":"lb_rows_run_verified.jsonl: each record's `model` field (the extractor's call that wrote decisions[])"},"note":"a model's judgement from the paper's own tables that they report on this leaderboard, not a result: no value from a paper that was not placed is printed or carried here; papers counts each paper once over these leaderboards","backfill":"8,886 of 18,637 newer papers checked so far (733 more had nothing to place; 41 could not be read)","backfill_complete":false,"leaderboards":[]},"papers_note":"The archive never published its papers-using-dataset list; these are papers with a leaderboard row on this dataset's benchmarks.","record_sha256":"36d6d6862b9f37681f7d358e60d2fec3fe3e8297e2f98d297d611dae0c390338","record_changed_at":"2026-10-05","record_changed_at_basis":"changed"}