Find research datasets worth reusing
Search datasets from major research repositories and use ShareScore to quickly assess how well each record supports discovery, access, and reuse.
166
datasets available to search
ShareScore release 0.7.1
Dataset results
166 results for “image classification”
Training CNNs with Low-Rank Filters for Efficient Image Classification: Trained Models
<p>Models from experiments referenced in the paper "Training CNNs with Low-Rank Filters for Efficient Image Classification", https://arxiv.org/abs/1511.06744</p> <p>Model names differ from those in the paper, but the csv files for each set of experiments relates the paper's name for the model and the real name of the model here:</p> <ul> <li>cifarma.csv: Network-in-Network CIFAR10 Models</li> <li>mitma.csv: MIT Places Models</li> <li>googlenetma.csv: GoogLeNet ILSVRC2012 Models</li> <li>vggma.csv: VGG-11 ILSVRC2012 Models</li> </ul> <p> </p> <p> </p>
Train and Evaluation Code, Road Classification Models and Test set of the paper "Impact of Image Resolution and Image Overlap on the Prediction Performance of Convolutional Neural Networks Trained for Road Classification"
<p>This repository contains the Python scripts built for training and evaluation of the implementation, together with the test data and the resulting road classification models corresponding to the paper "Impact of Image Resolution and Image Overlap on the Prediction Performance of Convolutional Neural Networks Trained for Road Classification". The scripts make use of the Tensorflow with Keras framework and the additional required dependencies.</p> <p>The training and validation set is based on the binary SROADEX dataset (https://zenodo.org/records/6482346) that was re-split into tiles that feature the image resolutions (256 x 256, 512 x 512, and 1024 x 1024 pixels) and image overlaps (0% and 12.5%) considered in this study. The data have been generated using scripts developed in Python using Open Source libraries (GDAL/OGR and MapScript) for rasterization of vector cartography that represents the axes of the different types of roads (urban, interurban and rural). This binary road data contains information from 16 full orthoimages (28.5 km * 18.5 km) with spatial resolution of 0.5 m/pixel from the insular and peninsular Spanish territory. Due to the size on disk of approximately 546 gigabytes, this training and validation data is only available upon request from the corresponding author. The test set has been generated from a novel area of 28.5 km * 18.5 km and features binary road labels. The test sets are provided in the repository for each resolution (with no overlap), so that additional DL models can be evaluated on the same data and compared with the results achieved in this study.</p> <p>The structure of the information shared in this repository is as follows:<br>The scripts have been grouped by tile resolution (256, 512 and 1024). First, the test set and the evaluation script can be found. For each tile resolution, there are two subfolders (corresponding to the "no overlap" and "12.5% overlap"). In each case, the Python scripts for training the models in the three repetitions are shared, and the trained models (H5 format) are shared in compressed form. Finally, for each resolution we also share the testing dataset which consists of two folders.</p> <p>The material is distributed under a CC-BY 4.0 license.</p>
→ Fig. 10. FESEM images of the test structure in lagenid foraminifers from Recent, Admiralty Bay, King George Island, West Antarctica (A) and from the Jurassic of Gnaszyn, Poland (B, C). A. Unilocular Procerolagena gracilis Williamson, 1848, MWGUW ZI/67/44/02. B. Unilocular Lagena globosa Montagu, 1803, MWGUW ZI/67/61/09. C. Uniserial Nodosaria pulchra Franke, 1936, MWGUW ZI/67/61/26. Oblique cross-sectional views (A1, A2, A4, B1, B2, C); transverse cross-sectional views, showing single-crystal interlocked bundle structures, inner pores which extend along the entire length of the bundles as well as prominent calcite cleavage (A3, B3). Abbreviations: c, prominent calcite cleavage; ip, inner pore. in Chamber arrangement versus wall structure in the high-rank phylogenetic classification of Foraminifera
→ Fig. 10. FESEM images of the test structure in lagenid foraminifers from Recent, Admiralty Bay, King George Island, West Antarctica (A) and from the Jurassic of Gnaszyn, Poland (B, C). A. Unilocular Procerolagena gracilis Williamson, 1848, MWGUW ZI/67/44/02. B. Unilocular Lagena globosa Montagu, 1803, MWGUW ZI/67/61/09. C. Uniserial Nodosaria pulchra Franke, 1936, MWGUW ZI/67/61/26. Oblique cross-sectional views (A1, A2, A4, B1, B2, C); transverse cross-sectional views, showing single-crystal interlocked bundle structures, inner pores which extend along the entire length of the bundles as well as prominent calcite cleavage (A3, B3). Abbreviations: c, prominent calcite cleavage; ip, inner pore.
Microscopy Imaging Dataset: Trypanosoma brucei Bloodstream Form Classification Using Deep Learning
<p>This dataset provides a comprehensive collection of microscopic images and associated labels, specifically designed to facilitate the automated classification of <em>Trypanosoma brucei</em> bloodstream forms—slender and stumpy. Accurate differentiation of these life cycle stages is vital for understanding the parasite's biology, transmission dynamics, and adaptation mechanisms in its mammalian host.</p> <p><strong>Contents:</strong></p> <ul> <li><strong>Image Data</strong>: Microscopic images of <em>T. brucei</em> bloodstream forms captured under standard imaging conditions, encompassing a broad array of image quality, cellular arrangements, and morphological characteristics.</li> <li><strong>Label Data</strong>: Annotation files for each image, specifying cellular forms as slender or stumpy, essential for supervised machine learning applications.</li> <li><strong>Supplementary Files</strong>: Additional Excel files providing information on training, testing, and validation splits, alongside test results for model evaluation.</li> </ul> <p><strong>Purpose:</strong></p> <p>This dataset serves as a valuable resource for researchers in parasitology, machine learning, and computational biology. It supports investigations into the biology and life cycle of <em>T. brucei</em>, while also providing a robust testbed for developing, validating, and benchmarking image processing and classification algorithms tailored to parasite morphology.</p> <p><strong>Data Collection and Methodology:</strong></p> <p>The dataset was compiled using advanced deep learning techniques, integrating the Cellpose segmentation algorithm with a custom-trained Xception model optimized for classifying <em>T. brucei</em> forms. The model achieved 97% classification accuracy, demonstrating effective application in handling complex cell images and distinguishing between slender and stumpy forms in challenging imaging conditions.</p> <p><strong>Usage:</strong></p> <p>Researchers are encouraged to use this dataset to:</p> <ul> <li>Analyze and classify the life cycle stages of <em>T. brucei</em> bloodstream forms in microscopic images.</li> <li>Develop and test deep learning models for single-cell image segmentation and classification.</li> <li>Explore cellular morphology patterns and refine machine learning approaches for other single-cell imaging applications.</li> </ul> <p><strong>Citation:</strong></p> <p>Please cite the original dataset if you utilize this resource in your research to acknowledge its contribution to the field.</p> <p><strong>Access and Availability:</strong></p> <p>This dataset is openly available through Zenodo, enabling researchers to download, explore, and apply it in various fields, from parasitology to advanced computational biology.</p>
Volunteer classifications of images from the Cropland Capture game
<p>Each entry represents a single classification of a single image by a volunteer rater.</p> <p>The dataset contains six columns:</p> <p>imgid: The unique identifier for each image used in the Cropland Capture campaign<br> userid: The unique identifier for each volunteer in the Cropland Capture campaign<br> rating: The answer provided; can be only one of the following:<br> 1: yes cropland<br> 2: no cropland<br> 0: maybe <br> date: Timestamp of the rating<br> ratingid: The unique identifier of the rating (this is different for each data row)<br> platform: What interface did the volunteer use to provide this rating?<br> 1: iPhone5<br> 2: iPhone, other models<br> 3: iPad<br> 4: Browser<br> >100: Android; different numbers indicate the screen size in pixels </p> <p>For more information, please see the following publications:</p> <p>Salk, CF, T Sturn, L See, S Fritz (2017). Limitations of majority agreement in crowdsourced image interpretation. <em>Transactions in GIS</em>, 21: 207–223.</p> <p>Salk, CF, T Sturn, L See, S Fritz (2016). Local knowledge and professional background have a minimal impact on volunteer citizen science performance in a land-cover classification task. <em>Remote Sensing</em>, 8: 744.</p> <p>Salk, CF, T Sturn, L See, S Fritz and C Perger (2016). Assessing quality of volunteer crowdsourcing contributions: Lessons from the Cropland Capture game. <em>International Journal of Digital Earth</em>, 9(4): 410-426.</p>
A tempοral Deep Convolutional Neural Network model on Sentinel-1 Image Time Series for pixel-wise Flood Classification (dataset)
<p>This is a dataset which has been designed to be used for flood time series classification. Each time series is annotated as flood or no-flood and represents a pixel-wise time series derived from stack of Sentinel-1 IW GRD images that have been pre-processed according to <a href="http://doi.org/10.5281/zenodo.6510223">https://doi.org/10.5281/zenodo.6510223</a>.</p>
Dataset for tumor infiltrating lymphocyte classification (304,097 image patches from TCGA)
<p>This is a dataset of images with or without tumor-infiltrating lymphocytes (TILs). The original images are from Abousamra et al. (2022) and Saltz et al. (2018), and the original whole slide images are from TCGA. This dataset is a subset of the data presented in Abousamra et al. (2022) (with new data partitions).</p> <p>If you use this dataset, please cite the following papers, as well as this Zenodo page.</p> <p>Abousamra, S., Gupta, M. D., Hou, L., Batiste, R., Zhao, T., Shankar, A., Rao, A., Chen, C., Samaras, D., Kurc, T., & Saltz, J. (2022). Deep Learning-Based Mapping of Tumor Infiltrating Lymphocytes in Whole Slide Images of 23 Types of Cancer. <em>Frontiers in Oncology</em>, 5971. https://doi.org/10.3389/fonc.2021.806603</p> <p>Saltz, J., Gupta, R., Hou, L., Kurc, T., Singh, P., Nguyen, V., Samaras, D., Shroyer, K. R., Zhao, T., Batiste, R., & Danilova, L. (2018). Spatial organization and molecular correlation of tumor-infiltrating lymphocytes using deep learning on pathology images. <em>Cell Reports</em>, <em>23</em>(1), 181-193.</p> <p> </p> <p>The acknowledgements from the <em>Frontiers in Oncology</em> and <em>Cell Reports</em> papers are included below:</p> <blockquote> <p>This work was supported by the National Institutes of Health (NIH) and National Cancer Institute (NCI) grants UH3-CA22502103, U24-CA21510904, 1U24CA180924-01A1, 3U24CA215109-02, and 1UG3CA225021-01 as well as generous private support from Bob Beals and Betsy Barton. AR and AS were partially supported by NCI grant R37-CA214955 (to AR), the University of Michigan (U-M) institutional research funds and also supported by ACS grant RSG-16-005-01 (to AR). AS was supported by the Biomedical Informatics & Data Science Training Grant (T32GM141746). This work was enabled by computational resources supported by National Science Foundation grant number ACI-1548562, providing access to the Bridges system, which is supported by NSF award number ACI-1445606, at the Pittsburgh Supercomputing Center, and also a DOE INCITE award joint with the MENNDL team at the Oak Ridge National Laboratory, providing access to Summit high performance computing system. The funders were not involved in the study design, collection, analysis, interpretation of data, the writing of this article or the decision to submit it for publication.</p> </blockquote> <p> </p> <blockquote> <p>We are grateful to all the patients and families who contributed to this study. Funding from the Cancer Research Institute is gratefully acknowledged, as is support from National Cancer Institute (NCI) through U54 HG003273, U54 HG003067, U54 HG003079, U24 CA143799, U24 CA143835, U24 CA143840, U24 CA143843, U24 CA143845,U24 CA143848, U24 CA143858, U24 CA143866, U24 CA143867, U24 CA143882, U24 CA143883, U24 CA144025, P30 CA016672, U24CA180924, U24CA210950, U24CA215109, NCI Contract HHSN261201400007C, and Leidos Biomedical Contract 14X138. A.U.K.R. and P.S were supported by CCSG Bioinformatics Shared Resource P30 CA01667, ITCR U24 Supplement 1U24CA199461-01, a gift from Agilent technologies, CPRIT RP150578, and a Research Scholar Grant from the American Cancer Society (RSG-16-005-01). This work used the Extreme Science and Engineering Discovery Environment (XSEDE), which is supported by National Science Foundation XSEDE Science Gateways program under grant ACI-1548562 allocation TG-ASC130023. The authors would like to thank Stony Brook Research Computing and Cyberinfrastructure and the Institute for Advanced Computational Science at Stony Brook University for access to the high-performance LIred and SeaWulf computing systems, the latter of which was supported by National Science Foundation grant (#1531492).</p> </blockquote> <p>------------------------------------</p> <p>This dataset includes 304,097 image patches. All images are 100 x 100 pixels at 0.5 micrometers per pixel. An image is TIL-positive if there are at least two TILs present.</p> <p>Refer to `images-tcga-tils-metadata.csv` for information about each image. That spreadsheet has the following columns:</p> <pre><code>partition,study,barcode,label,path,md5</code></pre> <p>Partition specifies which partition the image is part of (train, val, test). Study is the TCGA study the image is part of (e.g., acc for TCGA-ACC). Barcode is the TCGA participant barcode. This is used during partitioning, to ensure that images from the same participant are not present in different data partitions. Label is either til-negative or til-positive. An image is til-positive if there are at least two TILs in the image. Path is the path to the PNG image. All images are stored as PNG. Md5 is the md5 hash of the image. This can be used to ensure there are no duplicate images and to verify the integrity of images.</p> <p>There are study-specific directories in the directory `images-tcga-tils`, and there is a directory named `pancancer` that includes images from all the included TCGA studies. That directory uses symlinks to avoid storing duplicate data.</p> <p> </p>
MEDIC: A Multi-Task Learning Dataset for Disaster Image Classification
<p>Recent research in disaster informatics demonstrates a practical and important use case of artificial intelligence to save human lives and suffering during natural disasters based on social media contents (text and images). While notable progress has been made using texts, research on exploiting the images remains relatively under-explored. To advance image-based approaches, we propose MEDIC\footnote{Available~at: \url{https://crisisnlp.qcri.org/medic/index.html}}, which is the largest social media image classification dataset for humanitarian response consisting of 71,198 images to address four different tasks in a multi-task learning setup. This is the first dataset of its kind: social media images, disaster response, and multi-task learning research. An important property of this dataset is its high potential to facilitate research on \textit{multi-task learning}, which recently receives much interest from the machine learning community and has shown remarkable results in terms of memory, inference speed, performance, and generalization capability. Therefore, the proposed dataset is an important resource for advancing image-based disaster management and multi-task machine learning research. <br> </p>
Chess piece dataset for image classification
<p>Chess piece dataset for image classification. Contains 4 different chess sets, 3 used for training and the remainder for validation purposes. Each chess piece from each set has been photograph by a static bird's eyes camera from each of the 64 squares that form a chess board. This way, each piece is seen from all its different angles. </p>
CNN for the classification of ICE-CAMERA images of Antarctic ice particles
<p>-The file 'ZENODO_FILES.rar' contains the GoogleNet Convolutional Neural Network (CNN) trained to classify pre-processed ICE-CAMERA images (224*224*3) into 14 classes. CNN was developed for (Mathworks) MATLAB® R2020b.</p> <p>-The ICE-CAMERA images used for training, validation and testing the CNN are also contained in specific folders.</p> <p> </p>
Brain Tumor MR Image Data Set For Machine Vision Approach for Brain Tumor Classification using Multi Features Dataset
<p>The uploaded dataset contains the brain tumor MRI dataset. The dataset has been collected form the Bahawal Victoria Hospital, Bahawalpur, Pakistan. This dataset is an authorized MRI brain tumor dataset. Is has been authorized from the expert Radiologists of the Bahawal Victoria Hospital <a href="https://www.qamc.edu.pk/administration/2">BVH</a>. The dataset consists of three brain tumor types, namely adenomas, meningioma and glioma. it is only for academic, educational and experimental purpose. no other usage will be owned or any liability will be accepted by the authors.</p>
Train and Evaluation Code, Road Classification Models and Test set of the paper "Insights into the Effects of Image Overlap and Image Size on Semantic Segmentation Models Trained for Road Surface Area Extraction from Aerial Orthophotography"
<p>This repository contains the Python scripts built for training and evaluation of the implementation, together with the test data and the resulting road segmentation models corresponding to the paper "Insights into the Effects of Image Overlap and Image Size on Semantic Segmentation Models Trained for Road Surface Area Extraction from Aerial Orthophotography". The scripts make use of the Tensorflow with Keras framework and their additional required dependencies.</p> <p>The training and validation set is based on the binary SROADEX dataset (<a href="../records/6482346">https://zenodo.org/records/6482346</a>) that was re-split into tiles that feature the image resolutions (256 x 256, 512 x 512, and 1024 x 1024 pixels) and image overlaps (0% and 12.5%) considered in this study. The data have been generated using scripts developed in Python using Open Source libraries (GDAL/OGR and MapScript) for rasterization of vector cartography that represents the axes of the different types of roads (urban, interurban and rural). This binary road data contains information from 16 full orthoimages (28.5 km * 18.5 km) with spatial resolution of 0.5 m/pixel from the insular and peninsular Spanish territory. Due to the size on disk of approximately 492 gigabytes, this training and validation data is only available upon request from the corresponding author. The test set has been generated from a novel area from Palencia (Spain) and features 18 million pixels labelled with the positive "Road" class. The test sets are provided in the repository for each resolution (with no overlap), so that additional DL models can be evaluated on the same data and compared with the results achieved in this study.</p> <p>The structure of the information shared in this repository is as follows:<br>The scripts have been grouped by tile resolution (256, 512 and 1024). First, the test set and the evaluation script can be found. For each tile resolution, there are two subfolders (corresponding to the "no overlap" and "12.5% overlap"). In each case, the Python scripts for training the models in the three repetitions are shared, and the trained models (H5 format) are shared in compressed form. Finally, for each resolution we also share the testing dataset which consists of two folders.</p> <p>The material is distributed under a CC-BY 4.0 license.</p>
BRAIN Journal-An Enhancement over Texture Feature Based Multiclass Image Classification under Unknown Noise-Figure 8. Performance in 1st Approach for three data set
<p>This work also deals with classification of multi class images under different constraints of<br> data set. The first experiment is carried out on images without noise, second with Gaussian noise<br> and filtered data set in third experiment. Performance of the classifier using statistical texture<br> features for two approaches are presented in the table 3. It is observed that performance n the first<br> experiment is best in the first data set i.e. data set without noise in both the approach, while the<br> performance is decreased if the same images are affected by Gaussian noise.</p>
BRAIN Journal-An Enhancement over Texture Feature Based Multiclass Image Classification under Unknown Noise-Figure 4. Principal stages of image classification system
<p>In computer vision, images or objects are recognised by machine going through two phases<br> shown in the figure 4. First the system is trained with features extracted from sample images in<br> training stage then they are tested on input images in testing stage. The performance of the classifier<br> depends on features extracted from the image. This research work is carried out in three different<br> experiments, first experiment is performed on data set containing the original images of sixteen<br> categories, noisy images are classified in second experiment, and third experiment detects the type<br> of noise affected the image followed by filtering through appropriate filter, then filtered images are<br> classified. The performance of each of the experiment is measured with two approaches. First<br> approach extracts the statistical texture features of the whole image, and the original image of size<br> 128x128 is divided into sixteen blocks of size 32x32 pixels in second approach. Then six statistical<br> texture features discussed in second section are extracted from each of the block producing 96<br> features from each of the images are used for training and testing stage.</p>
BRAIN Journal-An Enhancement over Texture Feature Based Multiclass Image Classification under Unknown Noise-Figure 1. Sample Images of sixteen categories
<p>An image is often corrupted by noise in its acquisition or transmission. Noise is any<br> undesired information that degrades the image and appears in images from a variety of sources.</p> <p>Basically, there are three standard noise models [17], which model the types of noise<br> encountered in most images; they are additive noise, multiplicative noise and impulse noise. In this<br> work we have considered the occurrence of additive noise. An image function is given by f (x, y)<br> where (x, y) is spatial coordinate and f is intensity at point(x, y). Let f (x, y) be the original image,<br> g(x, y) be the noisy version and η(x, y) be the noise function, which returns random values coming<br> from an arbitrary distribution.</p>
BRAIN Journal-An Enhancement over Texture Feature Based Multiclass Image Classification under Unknown Noise-Figure 7. Performance in experiment 3 for both approaches
<p>Classification of noisy images starts with detection of type of noise followed by appropriate<br> filtering operation. Then similar approaches are followed for feature extraction as discussed in<br> above experiments.</p>
BRAIN Journal-An Enhancement over Texture Feature Based Multiclass Image Classification under Unknown Noise-Figure 4. Feed Forward Back propagation Neural Network
<p>This BPNN provides a computationally efficient<br> method for changing the weights in feed forward network, with differentiable activation function<br> units, to learn a training set of input-output data. Being a gradient descent method it minimizes the<br> total squared error of the output computed by the net. The aim is to train the network to achieve a<br> balance between the ability to respond correctly to the input patterns that are used for training and<br> the ability to provide good response to the input that are similar. A typical back propagation<br> network of input layer, one hidden layer and output layer is shown in figure 4.</p>
BRAIN Journal-An Enhancement over Texture Feature Based Multiclass Image Classification under Unknown Noise-Figure 6. Performance in experiment 2 for both approaches
<p>Second experiment is performed on noisy data set with 50 images each class for training of<br> feed forward neural network, and 100 images each class are used for testing phase. This experiment<br> is also carried out with two approaches as discussed in first experiment.</p>
BRAIN Journal-An Enhancement over Texture Feature Based Multiclass Image Classification under Unknown Noise-Figure 9. Performance in 2nd Approach for three data set
<p>This work also deals with classification of multi class images under different constraints of<br> data set. The first experiment is carried out on images without noise, second with Gaussian noise<br> and filtered data set in third experiment. Performance of the classifier using statistical texture<br> features for two approaches are presented in the table 3. It is observed that performance n the first<br> experiment is best in the first data set i.e. data set without noise in both the approach, while the<br> performance is decreased if the same images are affected by Gaussian noise. This is because the<br> texture feature of the original images consists Gaussian pattern also. Filtering of the noise from the<br> second data set improves the result. The table also shows that feature extraction using blocking of<br> the image enhance the average classification rate in all the case.</p>
BRAIN Journal-An Enhancement over Texture Feature Based Multiclass Image Classification under Unknown Noise-Figure 5. Performance in experiment 1 for both approaches
<p>The first experiment is carried out on data set containing 16 class and 50 images of each<br> class for training and 100 images of each class for testing. Initially the original images are resized to<br> 128X128 pixels and six texture features are extracted from 800 training images in the first<br> approach. It produces a feature matrix of size 6X800. Similarly 6X1600 feature matrix is produced<br> for testing stage. Then a neural classifier is designed with ten hidden layers and 16 output layers.<br> Feature matrix of training images are used to train the neural network. After that they are tested on feature matrix of test image. In second approach, the original images are divided into 16 blocks of<br> size 32X32 pixel each block. So it produces 96 features (16 blocksX6 features) for each image and<br> feature matrix of 96X800 for training image and 96X1600 feature matrix for testing samples.</p>
ScienceDex guides
Understand access before you commit
These curated guides explain access requirements, typical timelines, costs, and reuse considerations for widely used research datasets.
Allen Brain Atlas
Allen Brain Atlas is an Allen Institute collection of brain map atlases, datasets, APIs, and analysis tools covering mouse, human, and non-human primate brain resources.
Annotated Behaviour and Observability Dataset (ABODe)
ABODe is a University of Edinburgh DataShare dataset for behavior classification in group-housed mice using home-cage video, identities, bounding boxes, ground-plate positions, and annotator labels.
DANDI Archive for NWB datasets
DANDI is a BRAIN Initiative archive for publishing and sharing neurophysiology data, including electrophysiology, optophysiology, and behavioral data packaged as NWB and related standards.
International Brain Laboratory public data
The International Brain Laboratory public data releases expose standardized mouse decision-making experiments, including Neuropixels recordings, widefield calcium imaging, behavior, and session metadata accessed through the ONE API.
OpenNeuro
OpenNeuro is a free, open platform for sharing neuroimaging datasets, with public search, dataset pages, and download paths for web, S3, DataLad, and the OpenNeuro CLI.