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”
Replication Package for "Compatibility Issues in Deep Learning Systems: Problems and Opportunities"
<p>This dataset contains scripts and data used to generate relevant results for this paper. Detailed information and procedure to reproduce our results are described in README.md. </p> <p>code</p> <p>This folder contains two Python scripts: soextractor.py is used to extract 3,072 high-quality StackOverflow (SO) posts and soextractor_tags.py is used to extract the number of posts for the tags on SO. For detailed data collection criteria, please refer to Section 3.1 of our paper.</p> <p>DL compatibility issues.xlsx</p> <p>This file provides all the collected 3,072 issues, in which each line indicates whether the issue is a DL compatibility issue. Among them, 352 are DL compatibility issues. We also provide information on the library, stage, symptom, type, solution, root cause, and exception type for the DL compatibility issues. For the type CORE-TPL, we also provide backward-incompatible or forward-incompatible as well as API evolution patterns. For detailed manual classification of DL compatibility issues, please refer to Section 3.2 of our paper.</p> <p>Tool Survey.xlsx </p> <p>This file includes all the papers collected from the three top SE conferences (i.e., ICSE, FSE, and ASE) in recent five years (18-22). Each line of each sheet provides the following information: (a) Title, (b) Year, (c) Conference, and (d) Type. For the detailed paper collection procedure, please refer to Section 5 of our paper.</p>
Data from: Label-free timing analysis of SiPM-based modularized detectors with physics-constrained deep learning
<div> <div> <p>Pulse timing is an important topic in nuclear instrumentation, with far-reaching applications from high energy physics to radiation imaging. While high-speed analog-to-digital converters become more and more developed and accessible, their potential uses and merits in nuclear detector signal processing are still uncertain, partially due to associated timing algorithms which are not fully understood and utilized.</p> <p>In the paper "Label-free timing analysis of SiPM-based modularized detectors with physics-constrained deep learning", we propose a novel method based on deep learning for timing analysis of modularized detectors without explicit needs of labelling event data. By taking advantage of the intrinsic time correlations, a label-free loss function with a specially designed regularizer is formed to supervise the training of neural networks towards a meaningful and accurate mapping function. We mathematically demonstrate the existence of the optimal function desired by the method, and give a systematic algorithm for training and calibration of the model. The proposed method is validated on <strong>two experimental datasets</strong> based on silicon photomultipliers (SiPM) as main transducers:</p> <ol> <li>In the toy experiment, we collect data from a pair of SiPM sensors from a common laser source. The neural network model achieves the single-channel time resolution of 8.8 ps and exhibits robustness against concept drift in the dataset. </li> <li>In the electromagnetic calorimeter experiment, we collect data from an eight-channel calorimeter module. Several neural network models (Fully-Connected, Convolutional Neural Network and Long Short Term Memory) are tested to show their conformance to the underlying physical constraint and to judge their performance against traditional methods. </li> </ol> <p>In total, the proposed method works well in either ideal or noisy experimental condition and recovers the time information from waveform samples successfully and precisely. <strong>The dataset in this repository serves as a basis for similar researches on timing performance of SiPM-based nuclear detectors, and on application of neural networks to typical signals of nuclear radiation detectors.</strong></p> </div> </div>
FIGURE 5 U in Automated segmentation of insect anatomy from micro-CT images using deep learning
FIGURE 5 U-Net implementation. The architecture of the used convolutional neural network (CNN) is an implementation of U-Net. It consists of two parts: two 3×3 convolutions followed by 2×2 max pooling and two 3×3 convolutions followed by 2×2 upconvolutions. Dropout was added to avoid overfitting. As a final step a 1×1 convolution is applied, resulting in an output map with two classes.
FIGURE 6 in Automated segmentation of insect anatomy from micro-CT images using deep learning
FIGURE 6 Network performance evaluation. High true positive rate (TPR) and low false positive rate (FPR) values for training (blue) and testing data (red) indicate the network's high generalizability.
FIGURE 10 in Automated segmentation of insect anatomy from micro-CT images using deep learning
FIGURE 10 Application of pipeline for other insect species. The brain textures of various insect species can be very similar to those of ants, facilitating the prediction by the network even without pretraining on specific insect brain scans. (a) Raw image of wasp head (original 1000 × 1000 px) and (b) its prediction without postprocessing (original 520 × 520 px), indicating satisfactory identification of the borders of the brain area. (c) 2D image of praying mantis head (520 × 520 px) and (d) the prediction of its brain area without postprocessing. Even though the network overpredicts some small pixel islands, it excludes from its prediction areas of the muscles, fibers, and cuticle.
FIGURE 1 in Automated segmentation of insect anatomy from micro-CT images using deep learning
FIGURE 1 Segmentation pipeline overview. (a) Specimens are placed in iodine for staining for 2 weeks and then placed in small vials containing 99% ethanol to prevent them from moving during scanning. (b) The computed tomography (CT) scanner acquires successive X-ray images of the stepwise rotating specimen, and, using a user-defined reference image, automatically reconstructs them to produce orthogonal cross-section stacks that are used for the volume reconstruction of the specimen. (c) Volume rendering for future morphological studies is performed using Amira software. (d) Semiautomated segmentation of the brain volume of each scan (in orange) using the watershed method in Amira. (e) Schematic representation of the U-Net architecture used as the core of the pipeline for the development of a fully automated brain segmentation method. (f) The acquired brain images are used for training after preprocessing augmentation and manual creation of masks. (g) The network's prediction (in yellow) is postprocessed for smoothing out overpredicted areas (in red).
FIGURE 2 in Automated segmentation of insect anatomy from micro-CT images using deep learning
FIGURE 2 Exemplar images of full-body scans from different ant species. Three-dimensional (3D) reconstructed microcomputed tomography (micro-CT) image of (a) Acromyrmex versicolor and (b) Atta texana worker specimens, using volume rendering in Amira. (c) 2D micro-CT full body image of the Atta texana specimen (original 1000 × 1000 px). The brain area is the area with the most uniform pixel density within the whole body in its stained state, which makes it easy to recognize in most high-quality scans.
FIGURE 3 in Automated segmentation of insect anatomy from micro-CT images using deep learning
FIGURE 3 Example of semiautomated brain image segmentation. The brain area (in orange) of an Atta texana ant specimen was segmented using the watershed method in Amira; the 1000 × 1000 × 1000 px 3D image was manually postprocessed by smoothing and cropping oversegmented areas.
FIGURE 9 in Automated segmentation of insect anatomy from micro-CT images using deep learning
FIGURE 9 Prediction of ganglia in the thorax. As the tissue texture in the image is similar to that of the brain, the network accurately predicts other areas of nervous tissue in the organism. The pixel island detection step isolates the brain, but without this step neural tissue can be isolated.
FIGURE 8 3D in Automated segmentation of insect anatomy from micro-CT images using deep learning
FIGURE 8 3D volume of ant brain reconstructed from 2D images (original 520 × 520 px) predicted by the algorithm. 3D reconstructed brain prediction of an Atta texana worker.
FIGURE 7 in Automated segmentation of insect anatomy from micro-CT images using deep learning
FIGURE 7 Pipeline performance demonstrated both for validation (top row) and testing (bottom row) sets. (a, d) Raw images of head of Acromyrmex versicolor and Carebara atoma ant specimens, cropped along the x-y axes. The manually segmented brain areas are indicated in blue. (b, e) Network predictions before postprocessing (in yellow). Areas in yellow dotted circles are pixel islands not connected to the brain area that were overpredicted. (c, f) Predictions after postprocessing (in red). The borders of the predicted areas show good agreement with the manual segmentation in both sets. Note that in overlapping manually and automatically segmented areas in b, c, e, and f, colors appear green or purple.
Polymer Electrolyte Membrane Water Electrolyzer Oxygen Bubble Evolution Optical Video Recording For Deep Learning-Enhanced Characterization of Bubble Dynamics in Proton Exchange Membrane Water Electrolyzer by André Colliard-Granero, Keusra A. Gompou, Christian Rodenbücher, Kourosh Malek, Michael H. Eikerling, and Mohammad J. Eslamibidgoli
<p>Dataset used for the training of the segmentation model employed in the work "Deep Learning-Enhanced Characterization of Bubble Dynamics in Proton Exchange Membrane Water Electrolyzer" by André Colliard-Granero, Keusra A. Gompou, Christian Rodenbücher, Kourosh Malek, Michael H. Eikerling, and Mohammad J. Eslamibidgoli. This dataset consists in 35 images and the corresponding manual annotated masks of diverse bubbly scenarios extracted from the optical video recording of a PEMWE with a transparent flow field.</p>
Generative deep learning for hydrological forecasting: CVAE-75 basins from CANOPEX_v1
<p>Data associated with https://doi.org/10.1016/j.jhydrol.2023.130498</p>
Probabilistic volumetric speckle suppression in OCT using deep learning: Dataset
<p>This file contains a retinal OCT intensity volume as a demo dataset to generate volumetric speckle-suppressed training data using our non-local-means despeckling (TNode) script and four OCT intensity volumes and their corresponding TNode-processed intensity volumes of different tissue samples to train and test our deep learning framework used in "Probabilistic volumetric speckle suppression in OCT using deep learning" by Chintada et al. 2023. The TNode code for generating the training data and the source code for our deep learning framework are available at https://github.com/bhaskarachintada/DLTNode.git</p>
Data from: Deep learning-assisted near-Earth asteroid tracking in astronomical images
<p>This repository is the data release of our paper <em>Deep learning-assisted near-Earth asteroid tracking in astronomical images</em>. There are two categories in this repository:</p> <ul> <li>Simulated training dataset for training the star segmentation network. </li> </ul> <p>The dataset consists of two folders: image (grayscale images) and mask (binary images). The size of each image is 256*256.</p> <ul> <li>Example data for testing asteroid tracking algorithm.<br><br></li> </ul> <p>If you find this work useful, please cite our paper:</p> <div> <div>@article{du2024ASR,</div> <div>title = {Deep learning-assisted near-Earth asteroid tracking in astronomical images},</div> <div>journal = {Advances in Space Research},</div> <div>volume = {73},</div> <div>number = {10},</div> <div>pages = {5349-5362},</div> <div>year = {2024},</div> <div>issn = {0273-1177},</div> <div>doi = {https://doi.org/10.1016/j.asr.2024.02.048},</div> <div>url = {https://www.sciencedirect.com/science/article/pii/S0273117724001911},</div> <div>author = {Zhenhong Du and Hai Jiang and Xu Yang and Hao-Wen Cheng and Jing Liu},</div> <div>keywords = {Near-Earth asteroid, Deep learning, Convolutional neural network, Faint object extraction, Moving object linking},</div> <div>}</div> </div>
Fast Radio Map Estimation and Automated Radio Network Design Using Deep Learning App and Dataset
<p>In this research, we present the Deep Learning architecture encoder/fully-connected for estimate optimal desings WLANs in indoor scenarios. This architecture was implemented for WLAN structures consisting of 1, 2, 3, 4, and 5 access points, with the capability to perform the <a href="https://link.springer.com/chapter/10.1007/978-3-662-44415-3_4">Balanced k-means algortihm</a>, but in a fast manner.</p> <p><a href="https://github.com/johanflorez98/Fast-Radio-Map-Estimation-and-Automated-Radio-Network-Design-Using-Deep-Learning-App-and-Dataset/blob/main/README.md#general-dataset-structure">General Dataset Structure</a></p> <p>A major initial difficulty for starting the research was the lack of data, in this case, indoor scenario floor plans, users posisitons and optimal designs for training the architecture. Therefore, it was necessary to create an appropriate database that would facilitate the respective trainings. The dataset was created in base the <a href="https://link.springer.com/chapter/10.1007/978-3-662-44415-3_4">Balanced k-means algortihm</a>. This implementation was carried out in the MATLAB software <a href="https://github.com/johanflorez98/WLANs-optimal-designs-methodology">WLANs-optimal-designs-methodology</a>.</p> <p>Thus, this research provides a dataset that can be used for training multiple Deep Learning architectures and can facilitate future investigations into similar problems.</p> <p>We show a dataset composed by optimal designs for the 5GHz band WiFi in indoor scenarios: it has 102 indoor constructions plans and around 20 users distributions per plan floor as image and APs positions per case as coordinates. These distributions are random and several WLAN's structures: 1 to 5 access points.</p> <p>The above explain that we got a total of 61000 RME and CME, this presents that is a model without interference between channels.</p> <p>The pictures have a depth of 8 bits and size of <em>256pixels X 256pixels</em> equivalents to indoor constructions of <em>20 X 20</em> m<sup>2</sup>. These ones make reference to offices's spaces at general or classroom.</p> <p><a href="https://github.com/johanflorez98/Fast-Radio-Map-Estimation-and-Automated-Radio-Network-Design-Using-Deep-Learning-App-and-Dataset/blob/main/README.md#obtained-models">Obtained Models</a></p> <p>To evaluate the obtained models, the dataset consisting of floor plans 911 to 102 can be used, with available user distributions for each case of APs configurations as a test set, or other new images can be used.</p> <p>To manipulate the codes better, click here in the <a href="https://github.com/johanflorez98/Fast-Radio-Map-Estimation-and-Automated-Radio-Network-Design-Using-Deep-Learning-App-and-Dataset">repository</a>.</p>
Associated code and data for "A disease network-based deep learning approach for characterizing melanoma (doi:10.1002/IJC.33860)"
<p>This deposit contains the data, code, and analysis to recreate the results in the manuscript - Lai X, Zhou JF, Wissely A, Heppt M, Maier A, Berking C, Vera J, Zhang L. A disease network-based deep learning approach for characterizing melanoma. International Journal of Cancer. 2022; 150(6): 1029- 1044. <a href="http://www.researchgate.net/publication/355774213_A_disease_network-based_deep_learning_approach_for_characterizing_melanoma">doi:10.1002/IJC.33860</a>.</p> <p>If you have used the code for your research, please cite the original publication. Thank you very much.</p> <p> </p> <p> </p>
Deep learning reveals hotspots of global oceanic oxygen changes in the 21st century
<p>This is a global ocean-scale dissolved oxygen dataset derived from a deep learning method using hydrometeorological and biogeocheical driver factors. The dataset provides yearly estimates of dissolved oxygen concentration during 2003 to 2020 throughout the whole water column, with a spatial resolution of 0.25° × 0.25°. The unprecedented high-resolution dataset provides information on dissolved oxygen spatial distribution and temporal trend, making it a significant contribution to understanding global warming and anthropogenic stressors. </p>
Deep Learning Annotation Dataset and Images of Pea Aphids
<p><span>The small size and extensive polymorphisms of aphids make it difficult to identify larvae and adults solely based on their morphology. Here, we present an identification tool for the developmental stages of <em>Acyrthosiphon</em> <em>pisum</em> (Hemiptera: Aphididae) based on deep learning as a proof of concept. You Only Look Once (YOLO) algorithm is one of the most effective deep learning techniques for object detection. Although several studies have been conducted using deep learning technology for the detection and counting of tiny pests, the type of light source and size of the images were the limiting factors, as training was highly focused on uniform datasets and small insects. One way to overcome this problem is to introduce many types of datasets obtained from various light sources and microscopic magnifications. This strategy minimizes errors and omissions in aphid detection across all developmental stages in aphid individuals to the greatest extent possible. The experimental results showed that our modified YOLOv8 model could obtain over 95.9% and 99% accuracy for mean average precision (mAP) and Recall, respectively, under various light sources, such as yellow, white, and natural light, and stereomicroscope magnifications. This study showed an improved accuracy of aphid recognition at all developmental stages.</span><span> </span><span>The study presents a novel deep learning model utilizing the YOLO algorithm to identify developmental stages of </span><em><span>A</span></em><span>. </span><em><span>pisum</span></em><span>. This model achieves high accuracy across various light sources and magnifications, thereby enhancing aphid biology studies.</span></p>
dataset for "basic setting", "+ binary semantic loss", "+ class weights", "+ height weights", "+ region weights", "+ elastic distortion and subsampling", "+ TreeMix" in paper Automated forest inventory: analysis of high-density airborne LiDAR point clouds with 3D deep learning
<p>dataset for "basic setting", "+ binary semantic loss", "+ class weights", "+ height weights", "+ region weights", "+ elastic distortion and subsampling", "+ TreeMix" in paper Automated forest inventory: analysis of high-density airborne LiDAR point clouds with 3D deep learning</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.