The COCO (Common Objects in Context) dataset is a large-scale object detection, segmentation, and captioning dataset. It is designed to encourage research on a wide variety of object categories and is commonly used for benchmarking computer vision models. It is an essential dataset for researchers and developers working on object detection, segmentation, and pose estimation tasks.

Visual Genome contains Visual Question Answering data in a multi-choice setting. It consists of 101,174 images from MSCOCO with 1.7 million QA pairs, 17 questions per image on average. Compared to the Visual Question Answering dataset, Visual Genome represents a more balanced distribution over 6 question types: What, Where, When, Who, Why and How. The Visual Genome dataset also presents 108K images with densely annotated objects, attributes and relationships. Source: RaAM: A Relation-aware Attention Model for Visual Question Answering Image Source: Visual Genome: Connecting Language and Vision Using Crowdsourced Dense Image Annotations

PASCAL VOC 2007 is a dataset for image recognition. The twenty object classes that have been selected are: Person: person Animal: bird, cat, cow, dog, horse, sheep Vehicle: aeroplane, bicycle, boat, bus, car, motorbike, train Indoor: bottle, chair, dining table, potted plant, sofa, tv/monitor The dataset can be used for image classification and object detection tasks. Image Source: Object Detection and Recognition in Images

Click to add a brief description of the dataset (Markdown and LaTeX enabled). Provide: * a high-level explanation of the dataset characteristics * explain motivations and summary of its content * potential use cases of the dataset

BigEarthNet consists of 590,326 Sentinel-2 image patches, each of which is a section of i) 120x120 pixels for 10m bands; ii) 60x60 pixels for 20m bands; and iii) 20x20 pixels for 60m bands. Source: BigEarthNet: A Large-Scale Benchmark Archive For Remote Sensing Image Understanding

Our goal is to improve upon the status quo for designing image classification models trained in one domain that perform well on images from another domain. Complementing existing work in robustness testing, we introduce the first test dataset for this purpose which comes from an authentic use case where photographers wanted to learn about the content in their images. We built a new test set using 8,900 images taken by people who are blind for which we collected metadata to indicate the presence versus absence of 200 ImageNet object categories. We call this dataset VizWiz-Classification.

The COCO-MLT is created from MS COCO-2017, containing 1,909 images from 80 classes. The maximum of training number per class is 1,128 and the minimum is 6. We use the test set of COCO2017 with 5,000 for evaluation. The ratio of head, medium, and tail classes is 22:33:25 in COCO-MLT.

We construct the long-tailed version of VOC from its 2012 train-val set. It contains 1,142 images from 20 classes, with a maximum of 775 images per class and a minimum of 4 images per class. The ratio of head, medium, and tail classes after splitting is 6:6:8. We evaluate the performance on VOC2007 test set with 4952 images.

Sewer-ML is a sewer defect dataset. It contains 1.3 million images, from 75,618 videos collected from three Danish water utility companies over nine years. All videos have been annotated by licensed sewer inspectors following the Danish sewer inspection standard, Fotomanualen. This leads to consistent and reliable annotations, and a total of 17 annotated defect classes. Source: Sewer-ML Image Source: Sewer-ML

This dataset contains images of individual hand-written Bengali characters. Bengali characters (graphemes) are written by combining three components: a grapheme_root, vowel_diacritic, and consonant_diacritic. Your challenge is to classify the components of the grapheme in each image. There are roughly 10,000 possible graphemes, of which roughly 1,000 are represented in the training set. The test set includes some graphemes that do not exist in the train but has no new grapheme components. It takes a lot of volunteers filling out sheets like this to generate a useful amount of real data; focusing the problem on the grapheme components rather than on recognizing whole graphemes should make it possible to assemble a Bengali OCR system without handwriting samples for all 10,000 graphemes.

dacl10k stands for damage classification 10k images and is a multi-label semantic segmentation dataset for 19 classes (13 damages and 6 objects) present on bridges. The dacl10k dataset includes images collected during concrete bridge inspections acquired from databases at authorities and engineering offices, thus, it represents real-world scenarios. Concrete bridges represent the most common building type, besides steel, steel composite, and wooden bridges. 🏆 This dataset is used in the challenge associated with the "1st Workshop on Vision-Based Structural Inspections in Civil Engineering" at WACV2024.

LADI Overview The 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. Version 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, 2021, and 2022) 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". For 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: 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. ## Getting Started The LADI v2 dataset is available on Hugging Face at MITLL/LADI-v2-dataset. This is the recommended method. ### Default configuration The default configuration uses the v2a labels with images resized to fit within 1800x1200. This is the recommended configuration for most use cases. python from datasets import load_dataset ds = load_dataset("MITLL/LADI-v2-dataset") ### Advanced Usage If 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. The available configurations for the script are: v2, v2a, v2_resized, v2a_resized, v1_damage, and v1_infra. The 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. python from datasets import load_dataset ds = load_dataset("MITLL/LADI-v2-dataset", "v2a_resized", revision="script", streaming=True, download_ladi=True, base_dir='./ladi_dataset', trust_remote_code=True) This only needs to be done once, and subsequent calls can omit the argument, and it will automatically use your local copy. python ds = load_dataset("MITLL/LADI-v2-dataset", "v2a_resized", revision="script", streaming=True, base_dir='./ladi_dataset', trust_remote_code=True) You 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 version which resizes the images to 1800x1200, which should be large enough for most applications, but drastically reduces the overall dataset file size. The LADI v1 and v2 dataset files are hosted as part of the AWS Open Data program. ### Dataset Details The 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: - bridges_any - bridges_damage - buildings_affected - buildings_any - buildings_destroyed - buildings_major - buildings_minor - debris_any - flooding_any - flooding_structures - roads_any - roads_damage - trees_any - trees_damage - water_any The 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. #### Dataset v2a The 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. - bridges_any - buildings_any - buildings_affected_or_greater - buildings_minor_or_greater - debris_any - flooding_any - flooding_structures - roads_any - roads_damage - trees_any - trees_damage - water_any #### Supported Tasks The images are labeled for multi-label classification, as any number of the elements listed above may be present in a single image. #### Data Structure A single example in the v2a dataset looks like this: { 'image': <PIL.PngImagePlugin.PngImageFile image mode=RGB size=1800x1200 at ...>, 'bridges_any': False, 'buildings_any': False, 'buildings_affected_or_greater': False, 'buildings_minor_or_greater': False, 'debris_any': False, 'flooding_any': False, 'flooding_structures': False, 'roads_any': False, 'roads_damage': False, 'trees_any': True, 'trees_damage': True, 'water_any': True } ### Pretrained Classifiers We 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. - LADI-v2-classifier-small - Recommended for deployment, based on google/bit-50, trained on entire LADI v2 dataset - LADI-v2-classifier-large - Based on microsoft/swinv2, trained on entire LADI v2 dataset We 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. - LADI-v2-classifier-small-reference - LADI-v2-classifier-large-reference ## Citations ### LADI v2 If you use the LADI v2 dataset or classifiers, please cite the following: tex @misc{ladi_v2, title={LADI v2: Multi-label Dataset and Classifiers for Low-Altitude Disaster Imagery}, author={Samuel Scheele and Katherine Picchione and Jeffrey Liu}, year={2024}, eprint={2406.02780}, archivePrefix={arXiv}, primaryClass={cs.CV} } ## Distribution Information DISTRIBUTION STATEMENT A. Approved for public release. Distribution is unlimited. This 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. © 2024 Massachusetts Institute of Technology. The software/firmware is provided to you on an As-Is basis Delivered 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.