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.
1,273
datasets available to search
ShareScore release 0.9.0
Dataset results
1,273 results for “Deep Learning”
Predictors and predictands for "Downscaling CORDEX through deep learning to daily 1 km multivariate ensemble in complex terrain"
<p>Predictors and predictands for "Downscaling CORDEX through deep learning to daily 1 km multivariate ensemble in complex terrain". Training predictors from the ERA5 reanalysis, projecting predictors from CORDEX EUR11, and predictand from ReKIS (https://rekis.hydro.tu-dresden.de). Data is saved in ".rds" format, to be read from R, except for CORDEX files in NetCDF.</p>
Dataset for "Deep reinforcement learning for the olfactory search POMDP: a quantitative benchmark"
<p>Dataset containing the results shown in Fig. 6 of "Deep reinforcement learning for the olfactory search POMDP: a quantitative benchmark".</p>
Automatic Choroid Vascularity Index Calculation in Optical Coherence Tomography Images with Low Contrast Sclerocho-roidal Junction Using Deep Learning
<p>This project aims to calculate Choroid Vascularity Index (CVI) in optical coherenece tomography (OCT) images, using loss modified U-Net. The method is detailed in "Automatic Choroid Vascularity Index Calculation in Optical Coherence Tomography Images low contrast sclerochoroidal junction Using Deep Learning". The dataset consists of Enhanced-depth imaging optical coherence tomography images from two patient groups.</p> <p>• First dataset is including Raster OCT B-scans from patients with diabetic retinopathy.</p> <p>• Second dataset is including EDI-HD OCT B-scans from patients with pachychoroid spectrum.</p>
Electric Power Fuse Identification with Deep Learning
<p><strong>Paper Title</strong>: Electric Power Fuse Identification with Deep Learning</p> <p><strong>Journal Title</strong>: IEEE Transactions on Industrial Informatics</p> <p><strong>Accepted On</strong>: February 6, 2023</p> <p><strong>GitHub Link</strong>: <a href="https://github.com/MEDomics-UdeS/energAI-fuses">https://github.com/MEDomics-UdeS/energAI-fuses</a></p> <p><strong>Description</strong>: This project implements a supervised learning PyTorch-based end-to-end object detection pipeline for the purpose of detecting and classifying fuses in low-voltage electrical installations.</p> <p><strong>Files</strong>:</p> <ul> <li><strong>annotations.csv</strong>: Contains bounding box annotations for images. Save the file to 'data/annotations/' in the cloned <a href="https://github.com/MEDomics-UdeS/energAI-fuses">GitHub repository</a>.</li> <li><strong>final_model.pkl</strong>: Final trained model. Save the file anywhere and select it using the GUI from the cloned <a href="https://github.com/MEDomics-UdeS/energAI-fuses">GitHub repository</a> to use it.</li> <li><strong>images.zip</strong>: Images for learning and testing. Extract the .zip file to 'data/raw/' in the cloned <a href="https://github.com/MEDomics-UdeS/energAI-fuses">GitHub repository</a>.</li> </ul> <p><strong>Abstract</strong>:</p> <p>As part of arc flash studies, survey pictures of electrical installations need to be manually analyzed. A challenging task is to identify fuse types, which can be determined from physical characteristics such as shape, color and size. To automate this process using deep learning techniques, a new dataset of fuse pictures from past arc flash projects and data from the web was created. Multiple experiments were performed to train a final model, reaching an average precision (AP50) of 91.06 % on the holdout set, which confirms its potential for identification of fuse types in new photos. By identifying fuse types using physical characteristics only, the need to take clear pictures of the label text is eliminated, allowing pictures to be taken away from danger, thereby improving the safety of workers. All the resources needed to repeat the experiments are openly accessible, including the code and datasets.</p>
Deep Representation Learning of Physical Activity and Sleep Patterns During Pregnancy Identifies post-hoc Inferences Associated with Prematurity
<p><strong>Running title</strong>: series2signal gestational age "clock" for pregnancy monitoring</p> <p><strong>Summary</strong>: </p> <p>Preterm birth (PTB) is the leading cause of infant mortality globally. While research has focused on the development of predictive models for PTB, cost-effective interventions have remained understudied. Physical activity and sleep present unique opportunities for interventions in low- and middle-income populations. However, objective measurement of physical activity and sleep remains challenging and self-reported metrics suffer from low-resolution and accuracy that decays over time. In this study, we use physical activity data collected using a wearable device comprising over 181, 944 hours of data across N = 1, 083 patients. Using a new state-of-the art deep learning time-series classification architecture, we first develop a ”clock” of healthy dynamics in physical activity patterns during pregnancy by using gestational age (GA) as a surrogate for progression of pregnancy. We also developed a novel interpretability algorithm that integrates unsupervised clustering, model error analysis, feature attribution, and automated actigraphy analysis, allowing for model interpretation with respect to sleep, activity, and static clinical variables. Our model performs significantly better than 7 other machine learning and AI methods for modeling the progression of pregnancy based on measures of physical activity and sleep.</p> <p>Importantly, we found that deviations from this normal ”clock” of physical activity and sleep changes during pregnancy are strongly associated with pregnancy outcomes. When our model underestimates GA, there are 0.52 fewer preterm births than expected (P = 1.01e − 67) and when our model overestimates GA, there are 1.44 times (P = 2.82e − 39) more preterm births than expected. Model error is negatively correlated with interdaily stability (P = 0.043), indicating that our model assigns a more advanced GA when an individual’s daily rhythms are less precise. Supporting this, our model attributes higher importance to sleep periods in predicting higher-than-actual GA, relative to lower-than-actual GA (P = 1.01e − 21). Combining prediction with interpretability allows us to robustly signal when activity behaviors increase or decrease the likelihood of preterm birth and advocates for the future development of clinical decision support through passive monitoring and suggestions around exercise habits and sleep patterns, which are easily implemented in low- and middle-income countries (LMICs). Beyond this particular application, the presented pipeline can be used to analyze high-fidelity time-series data in other translational studies utilizing wearable devices.</p> <p> </p> <p><strong>Data description (brief)</strong>: the raw wearables data is available as .mtn files with the GA encoded in the filename after the underscore. The processed data with sleep annotations can be loaded using the pickle module for serialized objects in python. See https://github.com/nealgravindra/wearables for examples.</p>
Synthetic Nodules for Evaluation of 3D Deep Learning Segmentation Attribution
<p>Both the algorithm used to generate this dataset as well as the comprehensive evaluation metric for visual explanations are detailed in the paper "Attribution of 3D Deep Learning Segmentation in Medical Imaging".</p> <p>This is a synthetic dataset developed to enable the comprehensive evaluation of visual explanation methods applied to deep learning segmentation predictions. The data provided here include 100 training and 100 testing volumes, binary segmentation labels for model training, and explanation segmentation labels for explanation evaluation.</p> <p>The intended workflow for this dataset is: 1) Train a deep learning segmentation model using the 100 training volumes and binary segmentation labels. 2) Use a method of visual explanation to explain the trained model's segmentation decisions on the 100 testing volumes. 3) Use the explanation labels to evaluate the generated explanations.</p> <p> </p> <p>The folders <em>imagesTr</em> and <em>imagesTs</em> contain the training and testing image volumes, respectively. A <em>dataset.json</em> file has also been generated for the images to enable training using the nnUNet pipeline.</p> <p>The folders <em>labelsTr</em> and <em>labelsTs</em> contain the training and testing binary segmentation labels. In both cases, 1 indicates foreground voxels (spiculated nodule) and 0 indicates background voxels.</p> <p>The folders <em>labelsTr_full</em> and <em>labelsTs_full</em> contain the training and testing explanation labels. In both cases, 1 indicates non-spiculated nodule, 2 indicates spiculation structure (discriminating background), 3 indicates spiculated nodule body (segmentation foreground), and 0 indicates background.</p>
Deep Learning and Procrustes Analysis for Early Dysgraphia Risk Detection with a Tablet Application - DATASET
<p>Data to support the findings reported in the journal paper</p> <p>Lomurno, E., Dui, L. G., Gatto, M., Bollettino, M., Matteucci, M., & Ferrante, S. (2023). Deep Learning and Procrustes Analysis for Early Dysgraphia Risk Detection with a Tablet Application. <em>Life</em>, <em>13</em>(3), 598.</p> <ul> <li>Caratterizzazione_y3b_pulito.xls: <ul> <li>Subject ID</li> <li>Age at the beginning of the study</li> <li>Handedness</li> <li>Sex</li> </ul> </li> <li>Score_BVSCO_pulito.xls: <ul> <li>Subject ID</li> <li>columns B-D: graphemes in the three tests of the BVSCO-2</li> <li>columns E-G: z-score computed on normative data for the three tests of the BVSCO-2</li> <li>median z-score: computed on columns E-G</li> <li>at risk: if median z-score is under the -2 threshold (0=not at risk, 1=at risk)</li> </ul> </li> <li>executions.mat: a matrix for each time point (y1, y2a, y2b, y3a, y3b) and game (copa_q = copy square, copia_seq = copy sequence, tunnel_q = tunnel square, tunnel_ele = tunnel word). <ul> <li>first column of all the cell arrays: subject ID</li> <li>second column for the copy games: three sub-matrices with an execution for each level (spontaneous, big, small)</li> <li>columns 2:N for the tunnel games: an average execution for each index of difficulty</li> <li>for each execution, the columns of the sub-matrices are normalized time, normalized x, normalized y and pressure</li> </ul> </li> </ul> <p> </p> <p>Copy games</p> <ul> <li>A matrix for each game and each time point with a number of rows equal to the number of children and a number of columns equal to 2: <ul> <li>First column: array with school and child code</li> <li>Second column: executions</li> </ul> </li> <li>For the two copy games, the data has a dimension of 3 x 1, where 3 are the modalities (spontaneous, big, small) and 1 is the data matrix.</li> <li>By accessing the data matrix we have four variables, i.e., normalized time, normalized x, normalized y and pressure.</li> </ul> <p> </p> <p>Tunnel games</p> <ul> <li>A matrix for each game and each time point with a number of rows equal to the number of children and a number of columns equal to index of difficulty number +1: <ul> <li>First column: array with school and child code</li> <li>From the second column onwards: executions, each column corresponds to the "typical" execution (the average) of an ID (5 for square tunnel and 8 for ele tunnel). Sub-matrices contain the data matrix with its four variables. i.e., normalized time, normalized x, normalized y and pressure.</li> </ul> </li> </ul>
An intracochlear electrocochleography dataset: From raw data to objective analysis using deep learning
<p>Electrocochleography (ECochG) measures electrophysiological inner ear potentials in response to acoustic stimulation. These potentials reflect the state of the inner ear and provide important information about its residual function. For cochlear implant (CI) recipients, we can measure ECochG signals directly within the cochlea using the implant electrode. We are able to perform these recordings during and at any point after implantation.<br>However, the analysis and interpretation of ECochG signals are not trivial. To assist the scientific community, we provide our intracochlear ECochG data set, which consists of approximately 5,000 signals recorded from 46 ears with a cochlear implant. We collected data either immediately after electrode insertion or postoperatively in subjects with residual acoustic hearing. This data descriptor aims to provide the research community access to our comprehensive electrophysiological data set and algorithms. It includes all steps from raw data acquisition to signal processing and objective analysis using Deep Learning. In addition, we collected subject demographic data, hearing thresholds, subjective loudness levels, impedance telemetry, radiographic findings, and classification of ECochG signals.</p>
Datasets used to train the models in "Deep learning for denoising High-Rate Global Navigation Satellite System data."
<p>Datasets used to train the models in "Deep learning for denoising High-Rate Global Navigation Satellite System data." Additional information can be found at https://github.com/amtseismo/hrgnss_denoising.</p>
Single-jet datasets for particle reconstruction with deep learning
<p>Training and test datasets used for [1]* .</p> <ul> <li>singleQuarkJet_train.root : N=60649</li> <li>singleQuarkJet_test.root: N=38922</li> <li>singleGluonJet_test.root: N=38295</li> </ul> <p>The events are formed by a single initial state quark or gluon followed by parton shower generated in Pythia8 and then simulated using GEANT4 in a nearly-hermetic 6-layer calorimeter system as described in [1,2]. The branches in the ROOT files store features associated with cells, tracks, particles, pflow objects, jets, as well as edge lists for creating a graph representation of each event.<br> <br> *Note that subsets of N=50000 and N=30000 were used from the train and test samples, respectively, for the results in [1].</p> <p>[1] <a href="https://arxiv.org/abs/2212.01328">Reconstructing particles in jets using set transformer and hypergraph prediction networks</a><br> [2] <a href="https://arxiv.org/abs/2303.02101">Configurable Calorimeter Simulation for AI (COCOA)</a></p>
Towards Efficient Training in Deep Learning Side-Channel Attacks
<p>Datasets used to develop my Master's Thesis <em>Towards Efficient Training in Deep Learning Side-Channel Attacks</em> at Politecnico di Milano. </p> <p>The datasets contain power consumption measurements taken from multiple <em>Riscure Piñata </em>(STM32F4) boards (3) considering multiple keys (11) while executing AES-128.</p> <p>unprotected-AES.zip contains the traces related to the execution of a software unprotected implementation of AES-128.</p> <p>masked-AES.zip contains the traces related to the execution of a software masked implementation of AES-128.</p>
Manually labeled Bird song dataset of 22 species from Xeno-canto to enhance deep learning acoustic classifiers with contextual information.
<p>Data accompanying the paper: Jeantet and Dufourq (2023). Empowering Deep Learning Acoustic Classifiers with Human-like Ability to Utilize Contextual Information for Wildlife Monitoring. <em>Ecological Informatics</em>. 77, 15749541, DOI: 10.1016/j.ecoinf.2023.102256</p> <p> </p> <p>Our investigation contributes to the field of deep learning and bioacoustics by highlighting the potential for improved classification performance through the incorporation of contextual information such as time and location.</p> <p>To test if spatial-temporal information can enhance deep learning classifier, we developed a subset dataset derived from Xeno-Canto that included location metadata as input alongside the spectrogram. We used this dataset with the primary purpose of creating a bird song classification task with species carefully selected to share similar vocal characteristics but from distinct geographical distributions. We only considered the recordings of category `A', corresponding to the best quality score in the database.</p> <p>The dataset contains songs of <strong>22 bird species</strong> from 5 families and genera differents. The recordings were downloaded from the Xeno-canto database in .wav format and each recording was <strong>manually annotated </strong>by labelling the start and stop time for every vocalisation occurrence using Sonic Visualiser. In total, database contained 6537 occurrences of bird songs of various length from <strong>967 file recordings</strong>. A precise description of the distribution by species and country can be found in the associated article.</p> <p> </p> <p>The audio files are provided in "Audio.zip" and the manually verified annotation in "Annotations.zip". The name of each file follows the following nomenclature: Family_genus_species_country of recording_date of recording_ID Xenocanto_type of song.wav/svl. The meta-data information of each file can be find in the csv file provided (Xenocanto_metadata_qualityA_selection) based on the number of the ID Xeno-canto. The annotations can be viewed using the Sonic Visualiser software. The python codes to process these files and train neural networks can be found here : github</p> <p>The files were divided into a <strong>training folder</strong> and a<strong> validation folder</strong> to train and evaluate the efficiency of each method. For each species and country, we randomly selected 70% of the downloaded recordings for the training dataset and kept the remaining 30% for validation.</p> <p><strong>Process to select the species</strong> : We selected the ten most recorded families in the Passeriformes order, the most represented order in Xeno-canto database. From each of the ten families, we again sub-samples the ten most recorded genera. For each genus, we observed the countries of the recordings and the number of available recordings per species and countries. From these observations, we made a self-selection of genera containing species with similar songs but recorded in different regions, with enough recordings available by species and country to form a dataset . At the end, 5 genus were selected containing 22 species. We considered only recordings associated with bird songs, specifically, within Xeno-canto we selected the `song' type. To balance the number of recordings between species of the same genus, we reduced the number of recordings for the most represented species. Thus, for each genus we calculated the average of the number of records available per species and per country and limited the number of recordings for the species/country pairs that were in greater number to this value plus two.</p> <p> </p> <p> </p> <p> </p>
Deep learning models challenge the prevailing assumption that face-like effects for objects of expertise support domain-general mechanisms
<p>The question of whether perceptual expertise is mediated by general-expert or domain-specific processing mechanisms has been debated for decades. Because humans are experts in face recognition, face-like neural and cognitive effects for objects of expertise were considered to support for the general-expertise hypothesis. Conversely, stronger effects for faces than objects of expertise were considered to support the domain-specific hypothesis. However, the effects of domain, experience, and level of categorization, are confounded in human studies, which may lead to erroneous inferences. To overcome these limitations, we used computational models of perceptual expertise and tested different domains (objects, faces, birds) and levels of categorization (basic, sub-ordinate, individual) in isolation, matched for amount of experience. Like humans, the models generated a larger inversion effect for faces than for objects. Importantly, a face-like inversion effect was found for individual-based categorization of non-faces (birds) but only in a network specialized for that domain. Thus, contrary to prevalent assumptions, face-like effects in objects of expertise may originate from domain-specific rather than domain-general processing mechanisms. More generally, we show how deep learning algorithms can be used to isolate the effects of factors that are inherently confounded in the natural environment of biological organisms.</p>
Training patches and prediction codes of deep learning (LANA) model for Landsat 8/9 cloud/shadow mask
<p>This dataset includes (i) the image patches dataset and (ii) application/prediction (not training) codes for Landsat 8 cloud and cloud shadow masking used in a paper in review and uploaded here: </p> <p>Hankui Zhang, Dong Luo, David Roy, A learning attention network algorithm (LANA) for accurate Landsat-8 cloud and shadow masking, <em>Remote Sensing of Environment</em> </p> <p>The documentation is in <a href="https://zenodo.org/api/files/5462baa5-2bba-4b0f-92aa-c17681b6464b/l8_training_data_readme_new.pdf?versionId=95253efb-6447-46d8-ae4b-ac0c93b43532">l8_training_data_readme_new.pdf</a>. </p>
Dataset for the paper "An end-to-end deep learning framework for wideband signal recognition"
<p>This dataset comprises a collection of synthetically generated wideband signals, which were used in experiments conducted for the paper "An end-to-end deep learning framework for wideband signal recognition," submitted for publication to IEEE <em>Access</em>. In this work, the proposed learning-based approach for signal detection, localization, and classification was evaluated on the public wideband signal recognition dataset introduced by West <em>et al.</em> (<a href="https://ieeexplore.ieee.org/document/9593265">https://ieeexplore.ieee.org/document/9593265</a>). We note that the dataset provided here contains only additional auxiliary data that were generated to complement the public benchmark dataset in experiments that employ data mixing and transfer learning techniques. As such, the synthetic wideband signals in our dataset are designed to mimic those in the benchmark dataset.<br> <br> Specifically, in our simulations, a wideband signal is modeled as a superposition of several narrowband emissions and receiver noise. Similar to the benchmark dataset, all wideband signals have a normalized sampling frequency of <span class="math-tex">\(F_\text{s} = 1\)</span> sample per second (sps), duration of <span class="math-tex">\(L = 1 000 000\)</span> samples, and a center frequency of <span class="math-tex">\(F_\text{c} = 0\)</span> Hz. The narrowband signals have varying center frequencies, bandwidths, start times, and durations, and are modulated with a range of modulation classes, including:</p> <ol> <li>Amplitude Modulation - Double Sideband (AM-DSB)</li> <li>Amplitude Modulation - Single Sideband (AM-SSB)</li> <li>Frequency Modulation (FM)</li> <li><span class="math-tex">\(M\)</span>-Frequency-Shift Keying (<span class="math-tex">\(M\)</span>-FSK) for <span class="math-tex">\(M \in \{2, 4\}\)</span></li> <li><span class="math-tex">\(M\)</span>-Continuous Phase Frequency-Shift Keying (<span class="math-tex">\(M\)</span>-CPFSK) for <span class="math-tex">\(M \in \{2, 4\}\)</span></li> <li>Gaussian Minimum Shift Keying (GMSK)</li> <li>On-Off Keying (OOK)</li> <li><span class="math-tex">\(M\)</span>-Phase-Shift Keying (<span class="math-tex">\(M\)</span>-PSK) for <span class="math-tex">\(M \in \{2, 4, 8\}\)</span></li> <li><span class="math-tex">\(M\)</span>-Quadrature Amplitude Modulation (<span class="math-tex">\(M\)</span>-QAM) for <span class="math-tex">\(M \in \{16, 64, 256\}\)</span></li> </ol> <p>The modulation parameters for each narrowband signal are randomly selected from predefined typical ranges. For all wideband signals, the value of the power spectral density of the added white Gaussian noise has been initially set to -174 dBm/Hz. Further noise addition according to the desired signal-to-noise ratio (SNR) level can be applied by the user. <br> <br> The data are stored and documented according to the Signal Metadata Format (SigMF) standard (<a href="https://github.com/sigmf/SigMF">https://github.com/sigmf/SigMF</a>). Each wideband signal in the dataset is represented by a data file (a binary file containing digital I/Q samples of the signal) and a metadata file (in JSON format) that annotates the signal. The annotations include the general properties of the wideband recording (sampling rate, center frequency, signal duration, and noise power spectral density), as well as specific information for each narrowband signal (start time, duration, lowest and highest frequency, modulation class label, and power). For every signal in the dataset, the data file and the annotations file share the same name. For example, the first training wideband signal is stored in the file "train_1.sigmf-data" and is annotated by the file "train_1.sigmf-meta".</p>
Deep Reinforcement Learning for Data-Driven Adaptive Scanning in Ptychography
<p>These are the data sets used for the publication https://arxiv.org/abs/2203.15413.</p> <p>In more detail, the DataFile.pkl files in the training_data and testing_data folders are the actual data, including diffraction patterns, the corresponding reconstructions and the used illumination probe. The folders dwcnt_training_data and dwcnt_testing_data include the data sets of the simulated double-walled carbon nanotube.</p> <p>The reconstruction folder includes the source code of the reconstruction algorithm ROP and the used parameter files.</p> <p>The results folder includes the updated weights of the network model and the reconstructed potentials for the 25 test data sets used in the comparison of the publication.</p> <p>The sequence_grid.npy file stores the sparse grid scanning sequence used for initialization and comparison as described in the publication. The dwcnt_sequence_grid.npy file is the corresponding file for the double-walled carbon nanotube data.</p> <p>If you want more information, please contact the corresponding author of the publication, at schlozma@hu-berlin.de. </p>
DeepWeeds: A Multiclass Weed Species Image Dataset for Deep Learning
<p>DeepWeeds: A Multiclass Weed Species Image Dataset for Deep Learning</p> <p>This repository makes available the source code and public dataset for the work, "DeepWeeds: A Multiclass Weed Species Image Dataset for Deep Learning", published with open access by Scientific Reports: <a href="https://www.nature.com/articles/s41598-018-38343-3">https://www.nature.com/articles/s41598-018-38343-3</a>. The DeepWeeds dataset consists of 17,509 images capturing eight different weed species native to Australia in situ with neighbouring flora. In our work, the dataset was classified to an average accuracy of 95.7% with the ResNet50 deep convolutional neural network.</p> <p>The source code, images and annotations are licensed under <a href="https://creativecommons.org/licenses/by/4.0/">CC BY 4.0</a> license. The contents of this repository are released under an <a href="https://github.com/AlexOlsen/DeepWeeds/blob/master/LICENSE">Apache 2</a> license.</p> <p> </p> <p>Download the dataset images and our trained models</p> <ul> <li>images.zip (468 MB)</li> <li>models.zip (477 MB)</li> </ul> <p>Due to the size of the images and models they are hosted outside of the Github repository. The images and models must be downloaded into directories named "images" and "models", respectively, at the root of the repository. If you execute the python script (deepweeds.py), as instructed below, this step will be performed for you automatically.</p> <p> </p> <p><strong>TensorFlow Datasets</strong></p> <p>Alternatively, you can access the DeepWeeds dataset with <a href="https://www.tensorflow.org/datasets">TensorFlow Datasets</a>, TensorFlow's official collection of ready-to-use datasets. <a href="https://www.tensorflow.org/datasets/catalog/deep_weeds">DeepWeeds</a> was officially added to the TensorFlow Datasets catalog in August 2019.</p> <p> </p> <p><strong>Weeds and locations</strong></p> <p>The selected weed species are local to pastoral grasslands across the state of Queensland. They include: "Chinee apple", "Snake weed", "Lantana", "Prickly acacia", "Siam weed", "Parthenium", "Rubber vine" and "Parkinsonia". The images were collected from weed infestations at the following sites across Queensland: "Black River", "Charters Towers", "Cluden", "Douglas", "Hervey Range", "Kelso", "McKinlay" and "Paluma". The table and figure below break down the dataset by weed, location and geographical distribution.</p> <p> </p> <p> </p><p><strong>Data organization</strong></p> <p></p> <p>Images are assigned unique filenames that include the date/time the image was photographed and an ID number for the instrument which produced the image. The format is like so: <code>YYYYMMDD-HHMMSS-ID</code>, where the ID is simply an integer from 0 to 3. The unique filenames are strings of 17 characters, such as 20170320-093423-1.</p> <p> </p> <p><strong>labels</strong></p> <p>The labels.csv file assigns species labels to each image. It is a comma separated text file in the format:</p> <pre><code>Filename,Label,Species ... 20170207-154924-0,jpg,7,Snake weed 20170610-123859-1.jpg,1,Lantana 20180119-105722-1.jpg,8,Negative ... </code></pre> <p><em>Note: The specific label subsets of training (60%), validation (20%) and testing (20%) for the five-fold cross validation used in the paper are also provided here as CSV files in the same format as "labels.csv".</em></p> <p> </p> <p><strong>models</strong></p> <p>We provide the most successful ResNet50 and InceptionV3 models saved in Keras' hdf5 model format. The ResNet50 model, which provided the best results, has also been converted to UFF format in order to construct a TensorRT inference engine.</p> <pre><code>resnet.hdf5 inception.hdf5 resnet.uff </code></pre> <p> </p> <p><strong>deepweeds.py</strong></p> <p>This python script trains and evaluates Keras' base implementation of ResNet50 and InceptionV3 on the DeepWeeds dataset, pre-trained with ImageNet weights. The performance of the networks are cross validated for 5 folds. The final classification accuracy is taken to be the average across the five folds. Similarly, the final confusion matrix from the associated paper aggregates across the five independent folds. The script also provides the ability to measure the inference speeds within the TensorFlow environment.</p> <p>The script can be executed to carry out these computations using the following commands.</p> <ul> <li>To train and evaluate the ResNet50 model with five-fold cross validation, use <code>python3 deepweeds.py cross_validate --model resnet</code>.</li> <li>To train and evaluate the InceptionV3 model with five-fold cross validation, use <code>python3 deepweeds.py cross_validate --model inception</code>.</li> <li>To measure inference times for the ResNet50 model, use <code>python3 deepweeds.py inference --model models/resnet.hdf5</code>.</li> <li>To measure inference times for the InceptionV3 model, use <code>python3 deepweeds.py inference --model models/inception.hdf5</code>.</li> </ul> <p> </p> <p><strong>Dependencies</strong></p> <p>The required Python packages to execute deepweeds.py are listed in requirements.txt.</p> <p> </p> <p><strong>tensorrt</strong></p> <p>This folder includes C++ source code for creating and executing a ResNet50 TensorRT inference engine on an NVIDIA Jetson TX2 platform. To build and run on your Jetson TX2, execute the following commands:</p> <pre><code>cd tensorrt/src make -j4 cd ../bin ./resnet_inference </code></pre> <p> </p> <p><strong>Citations</strong></p> <p>If you use the DeepWeeds dataset in your work, please cite it as:</p> <p>IEEE style citation: “A. Olsen, D. A. Konovalov, B. Philippa, P. Ridd, J. C. Wood, J. Johns, W. Banks, B. Girgenti, O. Kenny, J. Whinney, B. Calvert, M. Rahimi Azghadi, and R. D. White, “DeepWeeds: A Multiclass Weed Species Image Dataset for Deep Learning,” <em>Scientific Reports</em>, vol. 9, no. 2058, <strong>2</strong> 2019. [Online]. Available: <a href="https://doi.org/10.1038/s41598-018-38343-3">https://doi.org/10.1038/s41598-018-38343-3</a> ”</p> <p> </p> <p><strong>BibTeX</strong></p> <pre><code>@article{DeepWeeds2019, author = {Alex Olsen and Dmitry A. Konovalov and Bronson Philippa and Peter Ridd and Jake C. Wood and Jamie Johns and Wesley Banks and Benjamin Girgenti and Owen Kenny and James Whinney and Brendan Calvert and Mostafa {Rahimi Azghadi} and Ronald D. White}, title = {{DeepWeeds: A Multiclass Weed Species Image Dataset for Deep Learning}}, journal = {Scientific Reports}, year = 2019, number = 2058, month = 2, volume = 9, issue = 1, day = 14, url = "https://doi.org/10.1038/s41598-018-38343-3", doi = "10.1038/s41598-018-38343-3" } </code></pre>
Deep Reinforcement Learning Enables Better Bias Control in Benchmark for Virtual Screening
<p>This compressed file contains all datasets made for the validation of MUBDsyn.</p><ul><li>datasets_int_val: 17 cases in this folder are derived from <a href="https://github.com/jwxia2014/ULS-UDS">MUBD for GPCRs</a>. MUBDreal was made by <a href="https://github.com/jwxia2014/MUBD-DecoyMaker2.0">MUBD-DecoyMaker2.0</a> and MUBDsyn was made by <a href="https://github.com/taoshen99/MUBDsyn">MUBD-DecoyMakersyn</a>.</li><li>datasets_ext_val_classical_VS: Five cases in this folder are derived from the shared cases of MUV and DUD-E. The active sets of MUV were taken as the input to make corresponding MUBD datasets. Files in SBVS are raw molecular docking results by smina.</li><li>datasets_ext_val_SI_classical_VS: DeepCoy and TocoDecoy were used to make the datasets corresponding to the same five cases above. The data of DeepCoy was directly retrieved from <a href="https://opig.stats.ox.ac.uk/resources">DeepCoy resources at OPIG</a> while topology decoys of TocoDecoy_9W were made based on the scripts provided at <a href="https://github.com/5AGE-zhang/TocoDecoy">TocoDecoy GitHub Repository</a>. Files in SBVS are raw molecular docking results by smina.</li><li>datasets_ext_val_ML_VS: Ten cases in this folder are derived from <a href="http://nrlist.drugdesign.fr/">NRLiSt-BDB</a>. Corresponding MUBD datasets were made as described above.</li></ul><p>All these datasets can be used for the reproduction of validation performed in the manuscript or to benchmark various virtual screening methods.</p>
Image dataset for cow identification, including code to train deep learning model, as well as analysis of results (SmARtview, 51088)
<p>This dataset and code was a result of the UKRI project "SmARtview: An AI-powered Augmented Reality Tool for Animal Health and Productivity", linked here: <a href="https://gtr.ukri.org/projects?ref=51088">https://gtr.ukri.org/projects?ref=51088</a></p> <p>These files are intended to be used for an accompanying publication in an academic journal.</p> <p>Anyone is free to use the contents for research and teaching purposes.</p>
Datasets for Deep Learning Based Radio Frequency Side-Channel Attack on Quantum Key Distribution
<p>The dataset contains measurements of radio-frequency electromagnetic emissions from a home-built sender module for BB84 quantum key distribution. The goal of these measurements was to evaluate information leakage through this side-channel. This dataset supplements our <a href="https://link.aps.org/doi/10.1103/PhysRevApplied.20.054040">publication</a> and allows to reproduce our results together with the source code hosted at <a href="https://github.com/XQP-Munich/EmissionSecurityQKD">GitHub</a> (and also on <a href="https://doi.org/10.5281/zenodo.7965628">Zenodo</a> via integration with GitHub).<br><br>The measurements are performed using a magnetic near-field probe, an amplifier and an oscilloscope. The dataset contains raw measured data in the file format output by the oscilloscope. Use our source code to make use of it. Detailed descriptions of measurement procedure can be found in our paper and in the metadata JSON files found within the dataset.</p> <p><strong>Commented list of datasets</strong></p> <p>This file lists the datasets that were analyzed and reported on in the paper. The datasets in the list refer to directories here. Note that most of the datasets contain additional files with metadata, which detail where and how the measurements were performed. The mentioned Jupyter notebooks refer to the source code repository https://github.com/XQP-Munich/EmissionSecurityQKD (not included in this dataset). Most of those notebooks output JSON files storing results. The processed JSON files are also included in the source code repository.</p> <p>In naming of datasets,</p> <ul> <li><em>Antenna</em> refers to the log-periodic dipole antenna. All datasets that do not contain `Antenna` in their name are recorded with the magnetic near-field probe.</li> <li><em>Rev1</em> refers to the initial electronics design, while `rev2` refers to the revised electronics design which contains countermeasures aiming to reduce emissions.</li> <li><em>Shielding</em> refers to measurements where the device is enclosed in a metallic shielding and the measurement takes place outside the shielding.</li> <li><em>Rotation</em> refers to orientation of the magnetic near-field probe at the same spacial location</li> </ul> <p><strong>Datasets collected with near-field probe for Rev1 electronics</strong></p> <ul> <li><strong>Rev1Distance</strong>: contains measurements at different distances from the Rev1 electronics performed above the FPGA. The deep learning attack is analyzed in `TEMPEST_ATTACK.ipynb`. The amplitude is analyzed in `get_raw_data_RMS_amplitude.ipynb`.</li> <li><strong>Rev12D</strong>: different locations on a 2d grid at a constant distance from the electronics. The deep learning attack is analyzed in `TEMPEST_ATTACK.ipynb`.</li> <li><strong>Rev130meas2.5cm</strong>: 30 measurements above the FPGA at a hight of 2.5cm. Used to evaluate how much amount of training data affects neural network performance. The deep learning attack is analyzed in notebooks `TEMPEST_ATTACK*.ipynb`. In particular, `TEMPEST_ATTACK_VARY_TRAINING_DATA.ipynb` is used on this dataset.</li> <li><strong>Rev1Rotation10deg</strong> contains a measurement for varying orientation of the probe at the same location. This is not mentioned in the paper and is only included for completeness. The deep learning attack is analyzed in notebooks `TEMPEST_ATTACK*.ipynb`.</li> <li><strong>Rev1TEMPESTShieldingFPGA</strong> Measurements with and without shielding at 4cm above the FPGA.</li> <li>- <strong>Rev1TEMPESTShieldingUSBHole</strong> Measurements with shielding in front of a hole of size about 2cm x 2cm. The deep learning attack is analyzed in `TEMPEST_ATTACK*.ipynb`.</li> </ul> <p><strong>Datasets collected with near-field probe for Rev2 electronics</strong></p> <ul> <li><strong>Rev2Distance</strong> contains measurements at different distances from the Rev2 electronics performed above the FPGA.</li> <li><strong>Rev22D</strong> and <strong>Rev22Dstart_7_0</strong> contain measurements on a 2d grid performed on the revised electronics. The dataset is split in two directories because the measurement procedure crashed in the middle. This split structure was kept in order to maintain consistency with the automatic metadata.</li> <li><strong>Rev230meas2.5cm</strong> 30 measurements above the FPGA at a hight of 2.5cm. Used to evaluate how much amount of training data affects neural network performance. The deep learning attack is analyzed in notebooks `TEMPEST_ATTACK*.ipynb`. In particular, `TEMPEST_ATTACK_VARY_TRAINING_DATA.ipynb` is used on this dataset.</li> </ul> <p><strong>Other datasets</strong></p> <ul> <li><strong>BackgroundTuesday</strong> background measurement (QKD device is not powered at all) performed with near-field probe on 2022 June 21st.</li> <li><strong>BackgroundSaturday</strong> background measurement (QKD device is not powered at all) performed with near-field probe on 2022 June 11th.</li> <li><strong>AntennaSpectra</strong> Dataset of spectra directly recorded by the oscilloscope. Used to demonstrate ability of telling apart the situation of sending QKD key (standard operation) and having the device turned on but not sending any key at a distance. Analyzed in notebook `Comparing_KeyNokey_Measurements.ipynb`.</li> <li><strong>Rev2ShieldingAntenna</strong> Raw amplitude measurements with log-periodic dipole antenna on Rev2 electronics including shielding enclosure, collected at various distances. None of our attacks against this scenario were successful. The dataset represents a challenge to test more advanced attacks using improved data processing.</li> </ul> <p> </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.