Papers with Code 아카이브

Image Classification

태스크의 하위 분류와 논문·코드·평가·기법·데이터셋을 탐색하세요.

하위 태스크
33
평가표
177
보존된 평가 기록
2,906

Image Classification is a fundamental task in vision recognition that aims to understand and categorize an image as a whole under a specific label. Unlike object detection, which involves classification and location of multiple objects within an image, image classification typically pertains to single-object images. When the classification becomes highly detailed or reaches instance-level, it is often referred to as image retrieval, which also involves finding similar images in a large database.

Source: Metamorphic Testing for Object Detection Systems

데이터셋

출처에 같은 태스크 이름으로 연결된 데이터셋입니다. 동명 태스크가 포함될 수 있으므로, 비교할 때는 평가 결과의 조건도 확인하세요.

CiNAT-Birds-2021

CiNAT Birds 2021 (Cross-View iNaturalist-2021 Birds) dataset contains ground-level images of bird species along with satellite images associated with the geolocation of the ground-level images. In total, there are 413,959 pairs for training and 14,831 pairs for validation and testing. The ground-level images are of varying sizes while the satellite images are of size 256x256. Additionally, the dataset comes with rich metadata for each image - geolocation, date, observer id, taxonomy.

CleanSTL-10

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

CNFOOD-241

Contains a dataset of 241 Chinese dishes with 191,811 images. There are 170843 images in the training set and 20943 images in the validation set. All images are resized to 600x600. As some of the images in the dataset are from ChineseFoodNet, they are not supported for commercial use.

CNFOOD-241-Chen

CNFOOD-241 Contains a dataset of 241 Chinese dishes with 191,811 images. There are 170843 images in the training set and 20943 images in the validation set. All images are resized to 600x600. As some of the images in the dataset are from ChineseFoodNet, they are not supported for commercial use. CNFOOD-241-Chen is the CNFOOD-241 dataset spilt with the list introduced in the paper "Res-VMamba: Fine-Grained Food Category Visual Classification Using Selective State Space Models with Deep Residual Learning," which has random split as train, val, test three parts.

COCO-N Medium

COCO-N Medium introduces a stochastic benchmark that simulates common real-world scenarios with noticeable label inaccuracies in the COCO dataset. This benchmark combines class and spatial noises to create a challenging yet realistic evaluation framework for instance segmentation models. It mimics datasets manually annotated by crowd workers, where a moderate level of label noise is expected. By incorporating both class and spatial inaccuracies, COCO-N Medium allows researchers to assess their models' basic robustness to label noise, providing insights into performance in typical real-world applications where perfect annotations are rare. This medium-level benchmark serves as a crucial middle ground, offering a more rigorous test than minimally noisy datasets while remaining within the bounds of commonly encountered data quality issues. COCO-N Medium enables a nuanced evaluation of model performance under realistic conditions, helping identify areas for improvement in handling noisy labels and guiding the development of more robust instance segmentation algorithms.

COCO-WAN (Medium noise)

The COCO-WAN benchmark is designed to assess the impact of weakly annotations (combined with auto-annotation tools) noise on instance segmentation models. This benchmark is built upon the COCO dataset and incorporates noise generated through weak annotations, simulating real-world scenarios where annotations might be imperfect due to semi-automated tools. It includes various levels of noise to challenge the robustness and generalization capabilities of segmentation models. Accurately labeling instance segmentation datasets is a complex and error-prone task, often leading to noisy labels. The COCO-WAN benchmark aims to provide a realistic testing ground for models to handle such noisy annotations. By utilizing foundation models and weak annotations, COCO-WAN simulates semi-automated annotation tools, helping researchers understand how well their models can perform under less-than-ideal labeling conditions. This benchmark includes multiple noise levels (easy, medium, and hard) to reflect varying degrees of annotation imperfections. Potential Use Cases of the Dataset: Model Robustness Testing: Researchers can use COCO-WAN to evaluate how different instance segmentation models cope with noisy annotations, allowing for the development of more resilient algorithms. Annotation Tool Improvement: By analyzing model performance on COCO-WAN, developers of annotation tools can identify common pitfalls and work on reducing noise in their outputs. Semi-Automated Annotation Systems: The benchmark provides insights into how models trained with semi-automated annotations perform, guiding improvements in such systems for better accuracy and efficiency in labeling tasks. The COCO-WAN benchmark offers a crucial resource for advancing the field of instance segmentation by highlighting the challenges posed by noisy labels and fostering the creation of more robust and reliable models. Model Robustness Testing: Researchers can use COCO-WAN to evaluate how different instance segmentation models cope with spatial, real noisy annotations, allowing for the development of more resilient algorithms. Semi-Automated Annotation Systems: The benchmark provides insights into how models trained with semi-automated annotations perform, guiding improvements in such systems for better accuracy and efficiency in labeling tasks. The COCO-WAN benchmark offers a crucial resource for advancing the field of instance segmentation by highlighting the challenges posed by noisy labels and fostering the creation of more robust and reliable models.

Cross-View Time Dataset (Cross-Camera Split)

The standard evaluation protocol of Cross-View Time dataset allows for certain cameras to be shared between training and testing sets. This protocol can emulate scenarios in which we need to verify the authenticity of images from a particular set of devices and locations. Considering the ubiquity of surveillance systems (CCTV) nowadays, this is a common scenario, especially for big cities and high visibility events (e.g., protests, musical concerts, terrorist attempts, sports events). In such cases, we can leverage the availability of historical photographs of that device and collect additional images from previous days, months, and years. This would allow the model to better capture the particularities of how time influences the appearance of that specific place, probably leading to a better verification accuracy. However, there might be cases in which data is originated from heterogeneous sources, such as social media. In this sense, it is essential that models are optimized on camera-disjoint sets to avoid learning sensor-specific characteristics that might not generalize accordingly for new imagery during inference. With this in mind, we propose a novel organization for CVT dataset. We split available data into training and testing sets, ensuring that all images from a single camera are assigned to the same set. During this division, we aimed to keep the size of each set roughly similar to the original splits, allowing models to be optimized with similar amounts of data.

Cultural Events Classification using Hyper-parameter Optimization of Deep Learning Technique

Abstract: Through digitization, maintaining and promoting cultural heritage is being strengthened. Concerning this background, this study presents a new Indonesia cultural events dataset and automatic image classification for cultural events. The dataset was developed using the Flickr image platform, and the five cultural events image was collected including the Baliem Festival, Jember Fashion Festival, Nyepi Festival, Pacu Jawi, and Pasola Festival. Further, Convolutional Neural Networks (CNN) was developed for the classification method. A comparison of CNN models (VGG16 and VGG19) using several optimization configurations was performed to get the best model. The results showed that the VGG16 with image augmentation and dropout regularization technique performed best with 94.66% accuracy. This study hoped to support the heritage's digital documentation process and preserve Indonesia's cultural heritage. * Image Source: The Flickr images under Attribution-Noncommercial License * Dataset Source: Indonesia Cultural Events Dataset * Paper Source: Link * The dataset consisted of 1,500 images and was divided into 300 images in each class. * Citation: Zhipeng, F., & Gani, H. (2021). Cultural Events Classification using Hyper-parameter Optimization of Deep Learning Technique. International Journal of Advanced Computer Science and Applications, 12(5). [APA Style]

DalleStreet

A dataset of images obtained from DALL-E 3 for 67 countries and 10 concept classes, similar to DollarStreet images.

DRIFT

The DRIFT dataset includes 25k image patches collected in five European countries sourced from aerial and nanosatellite image archives. Each image patch is associated with three target variables to predict: 1. Canopy height: average height value for pixels containing woody vegetation. 2. Tree count: number of overstory (visible from an overhead perspective) trees in the images. 3. Tree cover fraction: percentage of the image being covered by overstory tree crowns. The DRIFT dataset includes significant shifts between label and visual distributions due to sensor and area differences. Furthermore, vegetation tends to grow to fit the local climate, therefore introducing concept drift in the data: same tree species may appear differently in different subsets. The label distribution also varies among different subsets (countries). The dataset is a good choice for: * image-level regression * domain adaption for regression * remote sensing for forest applications

Dry Bean Dataset

Seven different types of dry beans were used in this research, taking into account the features such as form, shape, type, and structure by the market situation. A computer vision system was developed to distinguish seven different registered varieties of dry beans with similar features in order to obtain uniform seed classification. For the classification model, images of 13,611 grains of 7 different registered dry beans were taken with a high-resolution camera. Bean images obtained by computer vision system were subjected to segmentation and feature extraction stages, and a total of 16 features; 12 dimensions and 4 shape forms, were obtained from the grains. Source: UCI Machine Learning Repository

EarlyNSD

Early detection of plant nutritional deficiencies, followed by corrective actions, is essential for sustaining crop yield. However, identifying these early signs in plant leaves remains challenging, even with computer-aided diagnostic tools, due to their often subtle nature. To address this, we introduce a public dataset focused on three cucurbits: ash gourd, bitter gourd, and snake gourd. We chose to focus on these cucurbits due to their significant impact on global vegetable production. The dataset includes 2,700 segmented and augmented leaf images, capturing early indicators of nitrogen and potassium deficiencies alongside a healthy control group. This dataset, to the best of our knowledge, is the first to specifically target the critical early stage of nutritional stress in plants.

ENSeg

ENSeg Dataset Overview This dataset represents an enhanced subset of the ENS dataset. The ENS dataset comprises image samples extracted from the enteric nervous system (ENS) of male adult Wistar rats (Rattus norvegicus, albius variety), specifically from the jejunum, the second segment of the small intestine. The original dataset consists of two classes: - Control (C): Healthy animals. - Walker-256 Tumor (WT): Animals with cancer induced by the Walker-256 Tumor. Image acquisition involved 13 different animals, with 7 belonging to the C class and 6 to the WT class. Each animal contributed 32 image samples obtained from the myenteric plexus. All images were captured using the same setup and configuration, stored in PNG format, with a spatial resolution of 1384 × 1036 pixels. The overall process of obtaining the images and performing the morphometric and quantitative analyses takes approximately 5 months. ### Dataset Annotations Our dataset version includes expert-annotated labels for 6 animals, tagged as 2C, 4C, 5C, 22WT, 23WT, and 28WT. The image labels were created by members of the same laboratory where the images originated: researchers from the Enteric Neural Plasticity Laboratory of the State University of Maringá (UEM). Annotations were generated using LabelMe and consist of polygons marking each neuron cell. To maintain labeling quality according to laboratory standards, only neuron cells with well-defined borders were included in the final label masks. The labeling process lasted 9 months (from November 2023 to July 2024) and was iteratively reviewed by the lead researcher of the lab. ### Dataset Statistics After processing, the full dataset contains: - 187 images - 9,709 annotated neuron cells The table below summarizes the number of images and annotated neurons per animal tag. | Animal Tag | # of Images | # of Neurons | |------------|------------|--------------| | 2C | 32 | 1590 | | 4C | 31 | 1513 | | 5C | 31 | 2211 | | 22WT | 31 | 1386 | | 23WT | 31 | 1520 | | 28WT | 31 | 1489 | | Total | 187 | 9709 | ### Recommended Training Methodology Due to the limited number of animal samples and the natural split of images, we recommend using a leave-one-out cross-validation (LOO-CV) method for training. This ensures more reliable results by performing six training sessions per experiment, where: - In each session, images from one subject are isolated for testing. - The remaining images are used for training. Although cross-validation is recommended, it is not mandatory. Future works may introduce new training methodologies as the number of annotated subjects increases. However, it is crucial to maintain images from the same source (animal) within the same data split to prevent biased results. Even though samples are randomly selected from the animals' original tissues, this approach enhances the credibility of the findings. This dataset provides a valuable resource for instance segmentation and biomedical image analysis, supporting research on ENS morphology and cancer effects. Contributions and feedback are welcome! ## Project Citation If you want to cite our article, the dataset, or the source codes contained in this repository, please used the citation (bibtex format): @Article{felipe25enseg, AUTHOR = { Felipe, Gustavo Zanoni and Nanni, Loris and Garcia, Isadora Goulart and Zanoni, Jacqueline Nelisis and Costa, Yandre Maldonado e Gomes da}, TITLE = {ENSeg: A Novel Dataset and Method for the Segmentation of Enteric Neuron Cells on Microscopy Images}, JOURNAL = {Applied Sciences}, VOLUME = {15}, YEAR = {2025}, NUMBER = {3}, ARTICLE-NUMBER = {1046}, URL = {https://www.mdpi.com/2076-3417/15/3/1046}, ISSN = {2076-3417}, DOI = {10.3390/app15031046} } ## Additional Notes Please check out our previous datasets if you are interest into developing projects with Enteric Nervous System images: 1. EGC-Z: three datasets of Enteric Glial cells images, composed of three different chronic degenerative diseases: Cancer, Diabetes Mellitus, and Rheumatoid Arthritis. Each dataset represent binary classification task, with the classes: control (healthy) and sick; 2. ENS: the ENS image datasets comprises 1248 images taken from thirteen rats distributed in two classes: control/healthy or sick. The images were created with three distinct contrast settings targeting different Enteric Nervous System cells: Enteric Neuron cells, Enteric Glial cells, or both. For more details, please contact the main author of this project or create an issue on the project.

FathomNet2023

The FathomNet2023 competition dataset is a subset of the broader FathomNet marine image repository. The training and test images for the competition were all collected in the Monterey Bay Area between the surface and 1300 meters depth by the Monterey Bay Aquarium Research Institute. The images contain bounding box annotations of 290 categories of bottom dwelling animals. The training and validation data are split across an 800 meter depth threshold: all training data is collected from 0-800 meters, evaluation data comes from the whole 0-1300 meter range. Since an organisms' habitat range is partially a function of depth, the species distributions in the two regions are overlapping but not identical. Test images are drawn from the same region but may come from above or below the depth horizon. The competition goal is to label the animals present in a given image (i.e. multi-label classification) and determine whether the image is out-of-sample.

FGSCM-52

The development of the remote sensing fine-grained ship classification field necessitates large-scale realistic fine-grained ship datasets. The FGSC-23 and FGSCR-42 datasets have been pivotal for practical applications, yet they exhibit limitations when assessing advanced classification methods. Specifically, FGSC-23 categorizes data into relatively broad classes, such as lumping all auxiliary ships into one category, which does not align with the granularity required in real-world scenarios. Additionally, the dataset's limited number of categories fails to satisfy the diverse needs of practical applications. FGSCR-42, while providing finer classifications within some broader categories, still lacks comprehensive subcategory coverage. Moreover, the dataset suffers from an imbalance in sample distribution, with six categories containing fewer than ten samples each, which can hinder effective model training. To overcome these limitations, we introduce the FGSCM-52 dataset, an enhanced version of FGSCR-42, designed to offer a more extensive collection of fine-grained remote sensing ship images. FGSCM-52 includes an additional ten categories and increases the number of samples across existing categories, thereby making it serve as a better testbed for fine-grained ship classification.The dataset features image sizes ranging from 50$\times$50 to 1,600$\times$1,700 pixels. Furthermore, FGSCM-52 is equipped with well-defined hierarchical labels, which facilitate a more nuanced understanding of ship classification.

Fruits Dataset for Classification

Fruits Dataset for Classification About Dataset (strawberries, peaches, pomegranates) Photo requirements: 1-White background 2-.jpg 3- Image size 300*300 The number of photos required is 250 photos of each fruit when it is fresh and 250 photos of each fruit when it is rotten. Total 1500 images Diverse Collection With a diverse collection of Product images, the files provides an excellent foundation for developing and testing machine learning models designed for image recognition and allocation. Each image is captured under different lighting conditions and backgrounds, offering a realistic challenge for algorithms to overcome. Real-World Applications The variability in the dataset ensures that models trained on it can generalize well to real-world scenarios, making them robust and reliable. The dataset includes common fruits such as apples, bananas, oranges, and strawberries, among others, allowing for comprehensive training and evaluation. Industry Use Cases One of the significant advantages of using the Fruits Dataset for Classification is its applicability in various fields such as agriculture, retail, and the food industry. In agriculture, it can help automate the process of fruit sorting and grading, enhancing efficiency and reducing labor costs. In retail, it can be used to develop automated checkout systems that accurately identify fruits, streamlining the purchasing process. Educational Value The dataset is also valuable for educational purposes, providing students and educators with a practical tool to learn and teach machine learning concepts. By working with this dataset, learners can gain hands-on experience in data preprocessing, model training, and evaluation. Conclusion The Fruits Dataset for Classification is a versatile and indispensable resource for advancing the field of image classification. Its diverse and high-quality images, coupled with practical applications, make it a go-to dataset for researchers, developers, and educators aiming to improve and innovate in machine learning and computer vision. This dataset is sourced from Kaggle.

Gaze-CIFAR-10

We construct Gaze-CIFAR-10, a gaze-augmented image dataset based on the standard CIFAR-10 benchmark, enhanced with human eye-tracking annotations collected using the HTC VIVE Pro Eye headset. The original CIFAR-10 dataset consists of 60,000 color images across 10 categories, each with a resolution of $32 \times 32$ pixels. To enable reliable human gaze tracking, all images are upsampled to $1024 \times 1024$ using the Real-ESRGAN model. For each image, we collect a sequence of 176 eye-gaze coordinates, normalized to the range $[0, 224]$ (with the lower-left corner of the image as the origin) to match the input resolution of Vision Transformers (ViTs). Gaze data were collected from 20 participants using an HTC VIVE Pro Eye headset. The dataset contains 50,000 training images and 10,000 test images, with corresponding synchronized gaze trajectories.

Herbarium 2022

The Herbarium 2022: Flora of North America is a part of a project of the New York Botanical Garden funded by the National Science Foundation to build tools to identify novel plant species around the world. The dataset strives to represent all known vascular plant taxa in North America, using images gathered from 60 different botanical institutions around the world. In botany, a ‘flora’ is a complete account of the plants found in a geographic region. The dichotomous keys and detailed descriptions of diagnostic morphological features contained within a flora are used by botanists to determine which names to apply to plant specimens. This year's competition dataset aims to encapsulate the flora of North America so that we can test the capability of artificial intelligence to replicate this traditional tool —a crucial first step to harnessing AI’s potential botanical applications. The Herbarium 2022: Flora of North America dataset comprises 1.05 M images of 15,501 vascular plants, constituting more than 90% of the taxa documented in North America. Our dataset is constrained to include only vascular land plants (lycophytes, ferns, gymnosperms, and flowering plants). Our dataset has a long-tail distribution. The number of images per taxon is as few as seven and as many as 100 images. Although more images are available, we capped the maximum number in an attempt to ensure sufficient but manageable training data size for competition participants.

HRPlanesV2

The HRPlanesv2 dataset contains 2120 VHR Google Earth images. To further improve experiment results, images of airports from many different regions with various uses (civil/military/joint) selected and labeled. A total of 14,335 aircrafts have been labelled. Each image is stored as a ".jpg" file of size 4800 x 2703 pixels and each label is stored as YOLO ".txt" format. Dataset has been split in three parts as 70% train, %20 validation and test. The aircrafts in the images in the train and validation datasets have a percentage of 80 or more in size. Link: https://github.com/dilsadunsal/HRPlanesv2-Data-Set

Icon645

Icon645 is a large-scale dataset of icon images that cover a wide range of objects: * 645,687 colored icons * 377 different icon classes These collected icon classes are frequently mentioned in the IconQA questions. In this work, we use the icon data to pre-train backbone networks on the icon classification task in order to extract semantic representations from abstract diagrams in IconQA. On top of pre-training encoders, the large-scale icon data could also contribute to open research on abstract aesthetics and symbolic visual understanding.

Id Pattern Dataset

After defining a taxonomy of the main stone deterioration patterns and anomalies, we selected 354 highly representative images of stone-built heritage, offering them a careful selection of labels to choose from.

idsprites

Easily generate simple continual learning benchmarks. Inspired by dSprites. > It allows for procedurally generating a virtually infinite progression of random two-dimensional shapes. [...] we generate each unique shape in every combination of orientation, scale, and position. Most importantly, by providing the ground truth values of individual factors of variation (FoVs)

ImageNet 50 samples per class

This ImageNet version contains only 50 training images per class while the original testing set remains unchanged. It is one of the datasets comprising the data-efficient image classification (DEIC) benchmark. It was proposed to challenge the generalization capabilities of modern image classifiers.

Imbalanced-MiniKinetics200

Imbalanced-MiniKinetics200 was proposed by "Minority-Oriented Vicinity Expansion with Attentive Aggregation for Video Long-Tailed Recognition" to evaluate varying scenarios of video long-tailed recognition. Similar to CIFAR-10/100-LT, it utilizes an imbalance factor to construct long-tailed variants of the MiniKinetics200 dataset. Imbalanced-MiniKinetics200 is a subset of Mini-Kinetics-200 consisting of 200 categories which is also a subset of Kinetics400. Both the raw frames and extracted features with ResNet50/101 are provided.

IMPACT Patent

It is a large-scale multimodal patent dataset with detailed captions for design patent figures. 💥 Our dataset includes half a million design patents comprising 3.61 million figures along with captions from patents granted by the United States Patent and Trademark Office USPTO over a 16-year period from 2007 to 2022. * Multimodality: We introduce a multimodal patent dataset that includes patent images, metadata, and detailed captions to support a variety of NLP, vision, and multimodal tasks. This dataset is also valuable for patent analysis tasks such as classification, retrieval, prior art searches, and design trend analysis. * Comprehensive dataset: We have compiled a collection of 435,101 patents spanning 16 years from U.S. design patent documents. This extensive collection includes a total of 3,609,805 drawing figures. Additionally, our dataset consists of eleven fields such as the title, patent ID, claims, date of publication, classification code, and extensive image-related information, including the number of images per patent and descriptions of the viewpoints. * Descriptive captions: To address the absence of descriptions about the designs, such as features and shapes, we generate elaborated captions by employing a vision-language model. It generates descriptive captions for the design figures, capturing details from the sketch. These captions, coupled with the images, enrich our dataset and becomes a valuable resource for advanced patent analysis and multimodal research applications.

Indian Party Symbol Dataset

There was no predefined dataset of party symbols to be usedas a benchmark. We curated a dataset from various nationaland regional websites owned by the ECI. The dataset consists of symbols (image files) of 49 National and State registered parties approved by the ECI. For each image of theoriginal party symbol, 18 different distortions and transformations were created as variations to the training data. Each image is of the dimension 180 x 180. The final labeled dataset consists of 931 images of party symbols with their corresponding party names as the labels.

Iran's Built Heritage Binary Image Classification Dataset

Iran's Built Heritage Binary Image Classification Dataset contains approximately 10,500 CHB images gathered from four different sources: i) The archives of Iran’s cultural heritage ministry ii) The author’s (M.B) personal archives iii) images captured on site by the authors (M.B) during the research process iv) pictures crawled from the Internet but kept it to a minimum as their distribution differed due to heavy edits and effects.

IRMA

This collection compiles anonymous radiographs, which have been arbitrarly selected from routine at the Department of Diagnostic Radiology, Aachen University of Technology (RWTH), Aachen, Germany. The imagery represents different ages, genders, view positions and pathologies. Therefore, image quality varies significantly. All images were downscaled to fit into a 512 x 512 bounding box maintaining the original aspect ratio. All images were classified according to the IRMA code. Based on this code, 193 categories were defined. For 12,677 images, these categories are provided. The remaining 1,733 images without code are used as test data for the ImageCLEFmed 2009 competition. - Training data: 12,677 radiographs with known categories - Class distribution: Distribution of classes in the training data. - Test data: 1,733 radiographs without classification.

ISBNet

ISBNet is a dataset of images of recyclables. It is hand collected by our group at the International School of Beijing. The trash in these images was gathered from trash bins around the school. ISBNet totals 889 images distributed across 5 classes: cans (74), landfill (410), paper (182), plastic (122), and tetra pak (101). The data acquisition process involved using a piece of black poster paper as a background; this would create enough contrast for trash belonging to the paper category. These pictures were taken with an iPhone 8 and an iPhone XS. We recorded the trash bin in which the piece of trash originated from and any trash generating landmarks nearby. Please refer to the paper (ThanosNet: A Novel Trash Classification Method Using Metadata) for more about the format of the metadata.

JAMBO

The JAMBO dataset contains 3290 underwater images of the seabed captured by an ROV in temperate waters in the Jammer Bay area off the North West coast of Jutland, Denmark. All the images have been annotated by six annotators to contain one of three classes: sand, stone, or bad. Each of the six annotators have labelled all the images (that is, six individual annotations are provided for each image), which allows for analyzing inter-annotator disagreement and uncertainty.

Labels

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

LADI v2

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.

LAOFIW Dataset

An ancestral origin database of 14,000 images of individuals from East Asia, the Indian subcontinent, sub-Saharan Africa, and Western Europe. Source: Turning a Blind Eye: Explicit Removal of Biases and Variation from Deep Neural Network Embeddings

Large Labelled Logo Dataset (L3D)

It is composed of around 770k of color 256x256 RGB images extracted from the European Union Intellectual Property Office (EUIPO) open registry. Each of them is associated to multiple labels that classify the figurative and textual elements that appear in the images. These annotations have been classified by the EUIPO evaluators using the Vienna classification, a hierarchical classification of figurative marks. We suggest it to be used for: 1. Unconditional trademark generation 2. Conditional trademark generation. 3. Multi-label logo classification (Vienna classification). 4. Optical Character Recognition. 5. Conditional trade 6. Image segmentation. 7. Image retrieval

출처: Papers with Code · CC BY-SA 4.0. 아카이브 자료와 새로 수집한 논문은 별도로 관리합니다.