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.
55
datasets available to search
ShareScore release 0.9.0
Dataset results
55 results for “Dataset augmentation”
LGBTQIAphobia dataset (augmented and balanced)
<p><strong>Name: LGBTQIAphobia_dataset_augmented_balanced</strong><br><strong>Description</strong>: Labeled dataset with phrases retrieved from different digital sources (X/twitter, Instagram, TikTok) containing diverse messages directed towards the LGBTQIA+ community. It has 1000 phrases classified as {Non-LGBTQIAphobic (0), LGBTQIAphobic (1)} . It is the balanced version of <strong>LGBTQIAphobia_dataset_augmented.</strong><br><strong>Language: </strong>Spanish <br><strong>Format: </strong>CSV (UTF-8)<br><strong>Structure:</strong> id; phrase; class {0,1}<br><strong>Purpose</strong>: Be used for fine-tuned models that detect language offensive to Spanish or Latin LGBT communities in digital environments.<br><strong>Sources: </strong>X/Twitter, Instagram, TikTok, Youtube comments<br><strong>Size: </strong>20Kb <br><strong>Ethical considerations:</strong> This dataset was created strictly for academic and research purposes. We oppose any type of digital violence, in this case, against the LGBTQIA+ community. The person who was the target of the hate speech has been anonymised, and there is no intention to harm them in any way, either them or the person who delivered the speech. We prioritise the protection of the privacy and confidentiality of vulnerable individuals. To safeguard privacy, we carefully remove any identifying details, such as user IDs, phone numbers, and addresses, before sharing the data with our annotators. All the data we collect is from publicly available sources and does not contain any personal or sensitive information that may jeopardise anyone’s privacy. I request researchers to commit to abiding by ethical guidelines so as not to unnecessarily harm individuals.<br><strong>¿How was it created?<br></strong>- Starting recovery of discriminatory phrases for the LGBTQIA+ community from X/Twitter, Instagram, and Tiktok (197 phrases).<br>- Labelling by 3 raters as non-LGBTphobic (0) and LGBTphobic (1).<br>- Text augmentation was applied through backtranslation and random synonym replacement.<br>- Translating to Spanish part of <strong>McGiff, J., & Nikolov, N. S. (2024)</strong> dataset and was added under licence <strong>CC-BY-4.0<br>- </strong>To balance the majority class, we applied the undersampling technique.<strong><br></strong>- Finally, we obtained 1000 tagged phrases for version 1.0.2 of LGBTQIAphobia_augmented_balanced</p> <p><strong>Class distribution</strong><strong> </strong></p> <table style="border-collapse: collapse; height: 1%; width: 1%; border-width: 1px;"><colgroup><col style="width: 44.9675%;"><col style="width: 54.9128%;"></colgroup> <tbody> <tr style="height: 47.5868px;"> <td style="height: 47.5868px;"> <pre><strong>class</strong></pre> </td> <td style="height: 47.5868px;"> <pre><strong> instances</strong></pre> </td> </tr> <tr style="height: 24.5868px;"> <td style="height: 24.5868px;"> <pre><strong>0</strong></pre> </td> <td style="height: 24.5868px;"> <pre>513</pre> </td> </tr> <tr style="height: 47.5868px;"> <td style="height: 47.5868px;"> <pre><strong>1</strong></pre> </td> <td style="height: 47.5868px;"> <pre>487</pre> </td> </tr> </tbody> </table> <pre><strong>where class is</strong><br>0: non-lgbtphobic<br>1: lgbtphobic<br><br></pre>
Wallhack1.8k Dataset | Data Augmentation Techniques for Cross-Domain WiFi CSI-Based Human Activity Recognition
<p>This repository contains the <strong>Wallhack1.8k dataset</strong> for WiFi-based long-range activity recognition in Line-of-Sight (LoS) and Non-Line-of-Sight (NLoS)/Through-Wall scenarios, as proposed in [1,2], as well as the <strong>CAD models</strong> (of 3D-printable parts) of the WiFi systems proposed in [2].</p> <p><strong>PyTroch Dataloader</strong></p> <p>A minimal PyTorch dataloader for the Wallhack1.8k dataset is provided at: <a href="https://github.com/StrohmayerJ/wallhack1.8k" target="_blank" rel="noopener">https://github.com/StrohmayerJ/wallhack1.8k</a></p> <p><strong>Dataset Description</strong></p> <p>The Wallhack1.8k dataset comprises 1,806 CSI amplitude spectrograms (and raw WiFi packet time series) corresponding to three activity classes: "no presence," "walking," and "walking + arm-waving." WiFi packets were transmitted at a frequency of 100 Hz, and each spectrogram captures a temporal context of approximately 4 seconds (400 WiFi packets).</p> <p>To assess cross-scenario and cross-system generalization, WiFi packet sequences were collected in LoS and through-wall (NLoS) scenarios, utilizing two different WiFi systems (BQ: biquad antenna and PIFA: printed inverted-F antenna). The dataset is structured accordingly:</p> <ul> <li>LOS/BQ/ <- WiFi packets collected in the LoS scenario using the BQ system</li> <li>LOS/PIFA/ <- WiFi packets collected in the LoS scenario using the PIFA system</li> <li>NLOS/BQ/ <- WiFi packets collected in the NLoS scenario using the BQ system</li> <li>NLOS/PIFA/ <- WiFi packets collected in the NLoS scenario using the PIFA system</li> </ul> <p>These directories contain the raw WiFi packet time series (see Table 1). Each row represents a single WiFi packet with the complex CSI vector <em>H</em> being stored in the "data" field and the class label being stored in the "class" field. <em>H </em>is of the form [I, R, I, R, ..., I, R], where two consecutive entries represent imaginary and real parts of complex numbers (the Channel Frequency Responses of subcarriers). Taking the absolute value of <em>H</em> (e.g., via <em>numpy.abs(H)</em>) yields the subcarrier amplitudes <em>A</em>.</p> <p>To extract the 52 L-LTF subcarriers used in [1], the following indices of <em>A</em> are to be selected:</p> <pre><code># 52 L-LTF subcarriers csi_valid_subcarrier_index = [] csi_valid_subcarrier_index += [i for i in range(6, 32)] csi_valid_subcarrier_index += [i for i in range(33, 59)]</code></pre> <p>Additional 56 HT-LTF subcarriers can be selected via:</p> <pre><code># 56 HT-LTF subcarriers csi_valid_subcarrier_index += [i for i in range(66, 94)] csi_valid_subcarrier_index += [i for i in range(95, 123)]</code></pre> <p>For more details on subcarrier selection, see <a href="https://docs.espressif.com/projects/esp-idf/en/stable/esp32/api-guides/wifi.html">ESP-IDF</a> (Section Wi-Fi Channel State Information) and <a href="https://github.com/espressif/esp-csi">esp-csi</a>.</p> <p>Extracted amplitude spectrograms with the corresponding label files of the train/validation/test split: "trainLabels.csv," "validationLabels.csv," and "testLabels.csv," can be found in the <em>spectrograms/</em> directory.</p> <p>The columns in the label files correspond to the following: [Spectrogram index, Class label, Room label]</p> <ul> <li>Spectrogram index: [0, ..., n]</li> <li>Class label: [0,1,2], where 0 = "no presence", 1 = "walking", and 2 = "walking + arm-waving."</li> <li>Room label: [0,1,2,3,4,5], where labels 1-5 correspond to the room number in the NLoS scenario (see Fig. 3 in [1]). The label 0 corresponds to no room and is used for the "no presence" class.</li> </ul> <p><strong>Dataset Overview:</strong></p> <p>Table 1: Raw WiFi packet sequences.</p> <table> <tbody> <tr> <td><strong>Scenario</strong></td> <td><strong>System</strong></td> <td><em>"no presence" / label 0</em></td> <td><em>"walking" / label 1</em></td> <td><em>"walking + arm-waving" / label 2</em></td> <td><strong>Total</strong></td> </tr> <tr> <td>LoS</td> <td>BQ</td> <td>b1.csv</td> <td>w1.csv, w2.csv, w3.csv, w4.csv and w5.csv</td> <td>ww1.csv, ww2.csv, ww3.csv, ww4.csv and ww5.csv</td> <td> </td> </tr> <tr> <td>LoS</td> <td>PIFA</td> <td>b1.csv</td> <td>w1.csv, w2.csv, w3.csv, w4.csv and w5.csv</td> <td>ww1.csv, ww2.csv, ww3.csv, ww4.csv and ww5.csv</td> <td> </td> </tr> <tr> <td>NLoS</td> <td>BQ</td> <td>b1.csv</td> <td>w1.csv, w2.csv, w3.csv, w4.csv and w5.csv</td> <td>ww1.csv, ww2.csv, ww3.csv, ww4.csv and ww5.csv</td> <td> </td> </tr> <tr> <td>NLoS</td> <td>PIFA</td> <td>b1.csv</td> <td>w1.csv, w2.csv, w3.csv, w4.csv and w5.csv</td> <td>ww1.csv, ww2.csv, ww3.csv, ww4.csv and ww5.csv</td> <td> </td> </tr> <tr> <td> </td> <td> </td> <td>4</td> <td>20</td> <td>20</td> <td><strong>44</strong></td> </tr> </tbody> </table> <p>Table 2: Sample/Spectrogram distribution across activity classes in Wallhack1.8k.</p> <table> <tbody> <tr> <td><strong>Scenario</strong></td> <td><strong>System</strong></td> <td> <p><em>"no presence" / </em> label 0</p> </td> <td> <p><em>"walking"</em> / label 1</p> </td> <td><em>"walking + arm-waving" / </em>label 2</td> <td><strong>Total</strong></td> </tr> <tr> <td>LoS</td> <td>BQ</td> <td>149</td> <td>154</td> <td>155</td> <td> </td> </tr> <tr> <td>LoS</td> <td>PIFA</td> <td>149</td> <td>160</td> <td>152</td> <td> </td> </tr> <tr> <td>NLoS</td> <td>BQ</td> <td>148</td> <td>150</td> <td>152</td> <td> </td> </tr> <tr> <td>NLoS</td> <td>PIFA</td> <td>143</td> <td>147</td> <td>147</td> <td> </td> </tr> <tr> <td> </td> <td> </td> <td>589</td> <td>611</td> <td>606</td> <td><strong>1,806</strong></td> </tr> </tbody> </table> <p> </p> <p><strong>Download and Use</strong><br>This data may be used for non-commercial research purposes only. If you publish material based on this data, we request that you include a reference to one of our papers [1,2].</p> <p>[1] Strohmayer, Julian, and Martin Kampel. (2024). “Data Augmentation Techniques for Cross-Domain WiFi CSI-Based Human Activity Recognition”, <em>In IFIP International Conference on Artificial Intelligence Applications and Innovations</em> (pp. 42-56). Cham: Springer Nature Switzerland<em>,</em> doi: <a href="https://doi.org/10.1007/978-3-031-63211-2_4" target="_blank" rel="noopener">https://doi.org/10.1007/978-3-031-63211-2_4</a>.</p> <p>[2] Strohmayer, Julian, and Martin Kampel., “Directional Antenna Systems for Long-Range Through-Wall Human Activity Recognition,” <em>2024 IEEE International Conference on Image Processing (ICIP)</em>, Abu Dhabi, United Arab Emirates, 2024, pp. 3594-3599, doi: <a href="https://doi.org/10.1109/ICIP51287.2024.10647666" target="_blank" rel="noopener">https://doi.org/10.1109/ICIP51287.2024.10647666</a>.</p> <p>BibTeX citations:</p> <pre>@inproceedings{strohmayer2024data, title={Data Augmentation Techniques for Cross-Domain WiFi CSI-Based Human Activity Recognition}, author={Strohmayer, Julian and Kampel, Martin}, booktitle={IFIP International Conference on Artificial Intelligence Applications and Innovations}, pages={42--56}, year={2024}, organization={Springer}}<br><br>@INPROCEEDINGS{10647666,<br> author={Strohmayer, Julian and Kampel, Martin},<br> booktitle={2024 IEEE International Conference on Image Processing (ICIP)}, <br> title={Directional Antenna Systems for Long-Range Through-Wall Human Activity Recognition}, <br> year={2024},<br> volume={},<br> number={},<br> pages={3594-3599},<br> keywords={Visualization;Accuracy;System performance;Directional antennas;Directive antennas;Reflector antennas;Sensors;Human Activity Recognition;WiFi;Channel State Information;Through-Wall Sensing;ESP32},<br> doi={10.1109/ICIP51287.2024.10647666}}<br><br><br></pre>
ASVspoof2019LA-Sim: Augmented Dataset for An Empirical Study on Channel Effects for Synthetic Voice Spoofing Countermeasure Systems
<p>This is the dataset we augmented to study the channel effects for anti-spoofing. For more details, please refer to our Interspeech 2021 paper: "An Empirical Study on Channel Effects for Synthetic Voice Spoofing Countermeasure Systems".</p> <p>Proceeding: <a href="https://www.isca-speech.org/archive/interspeech_2021/zhang21ea_interspeech.html">https://www.isca-speech.org/archive/interspeech_2021/zhang21ea_interspeech.html</a></p> <p>Arxiv: <a href="https://arxiv.org/pdf/2104.01320.pdf">https://arxiv.org/pdf/2104.01320.pdf</a></p> <p>Code: <a href="https://github.com/yzyouzhang/Empirical-Channel-CM">https://github.com/yzyouzhang/Empirical-Channel-CM</a></p> <p>Contact: you.zhang@rochester.edu</p> <p><strong>Version 1.0</strong> contains the <strong>training</strong> and the <strong>development</strong> set. We have added the <strong>evaluation</strong> set in <strong>version 1.1 </strong>but deleted the training set due to the size limitation, but you can still access the training set in version 1.0.</p> <p>Please check it out.</p> <p>To extract the files, please use the following commands:</p> <pre><code class="language-bash">cat eval.tar.gz-part* > eval.tar.gz tar -xvzf *.tar.gz</code></pre> <p>After concatenation, to make sure the download is complete, you can check with the following:</p> <pre><code>md5sum *.tar.gz 15dea7d28b126994bb6b159778f706af dev.tar.gz 0615052b34ca6c7f58505eaa8647844f eval.tar.gz 3058dd9d407f3c9ae697acca8c34a6c3 train.tar.gz</code></pre> <p>Thanks.</p>
SEMFIRE forest dataset for semantic segmentation and data augmentation
<p><strong>SEMFIRE Datasets (Forest environment dataset)</strong></p> <p>These datasets are used for semantic segmentation and data augmentation and contain various forestry scenes. They were collected as part of the research work conducted by the Institute of Systems and Robotics, University of Coimbra <a href="https://isr.uc.pt/index.php/people?task=showprojects.show&idProject=203">team</a> within the scope of the Safety, Exploration and Maintenance of Forests with Ecological Robotics (SEMFIRE, ref. <a href="http://semfire.ingeniarius.pt/">CENTRO-01-0247-FEDER-032691</a>) research project coordinated by <a href="https://ingeniarius.pt/">Ingeniarius Ltd.</a></p> <p>The semantic segmentation algorithms attempt to identify various semantic classes (e.g. background, live flammable materials, trunks, canopies etc.) in the images of the datasets.</p> <p>The datasets include diverse image types, e.g. original camera images and their labeled images. In total the SEMFIRE datasets include about 1700 image pairs. Each dataset includes corresponding .bag files.</p> <p>To launch those .bag files on your ROS environment, use the instructions on the following Github <a href="https://github.com/Forestry-Robotics-UC/fruc_rosbags">repository</a></p> <p>Description of<strong> </strong>each <strong>dataset:</strong></p> <ol> <li><strong>2019_2020_quinta_do_bolao_coimbra:</strong> Robot moving on a path through a forest environment</li> <li><strong>2020_ctcv_parking_lot_coimbra:</strong> Robot moving in a circle in a parking lot for testings</li> <li><strong>2020_sete_fontes_forest: </strong>A set of forest images acquired by hand-held apparatus</li> </ol> <p>Each <strong>dataset</strong> consists of following <strong>directories:</strong></p> <ol> <li><strong>images directory: </strong>diverse image types, e.g. original camera images and their labeled images</li> <li><strong>rosbags directory: </strong>.bag files, which correspond to the image directory</li> </ol> <p>Each <strong>images directory </strong>consists of following <strong>directories:</strong></p> <ul> <li><strong>img:</strong> original camera images</li> <li><strong>lbl:</strong> single channel images (ground truth) with corresponding labels for each image in<strong> img</strong></li> <li><strong>lbl_colored: </strong>camera <strong> </strong>images in <strong>lbl</strong> colorized according to different semantic classes (for more details see the datasets descriptions)</li> <li><strong>lbl_overlaid: </strong>camera images in <strong>img </strong>overlaid with corresponding labels (colored)</li> </ul> <p>Each <strong>rosbags directory </strong>contains .bag files with the following <strong>topics:</strong></p> <ul> <li><strong>2019_2020_quinta_do_bolao_coimbra_rosbags: </strong> <ul> <li>/back_lslidar_packet</li> <li>/dalsa_camera_720p/compressed</li> <li>/flir_ax8/compressed</li> <li>/front_lslidar_packet</li> <li>/gps_fix</li> <li>/gps_time</li> <li>/gps_vel</li> <li>/imu/data</li> <li>/realsense/aligned_depth_to_color/image_raw</li> <li>/realsense/color/camera_info</li> <li>/realsense/color/image_raw/compressed</li> <li>/realsense/depth/camera_info</li> <li>/realsense/depth/image_rect_raw/compressed</li> <li>/realsense/extrinsics/depth_to_color</li> </ul> </li> <li><strong>2020_ctcv_parking_lot_coimbra_rosbags:</strong> <ul> <li>/dalsa_camera_720p/compressed</li> <li>/gps_fix</li> <li>/gps_ime</li> <li>/fused_point_cloud</li> <li>/imu/data</li> <li>/imu/mag</li> <li>/imu/rpy</li> </ul> </li> <li><strong>2020_sete_fontes_forest_rosbags: </strong> <ul> <li>/realsense/camera_info</li> <li>/realsense/depth_compressed/compressedDepth</li> <li>/realsense/nir/left/compressed</li> <li>/realsense/nir/right/compressed</li> <li>/realsense/rgb/compressed</li> </ul> </li> </ul> <p>All datasets include a detailed description as a text file. In addition, they include a rosbag_info.txt file with a description for each ROS inside the .bag files as well as a description for each ROS topic.</p> <p> </p> <p>The following table shows the statistical description of typical portuguese woodland configurations with structured plantations of <em>Pinus pinaster </em>(<em>Pp, </em>pine trees) and <em>Eucalyptus globulus </em>(<em>Eg, </em>eucalyptus).</p> <table> <tbody> <tr> <td> </td> <td><strong>"Low density" structured plantation</strong></td> <td><strong>"High density" structured plantation</strong></td> </tr> <tr> <td><strong>Tree density (assuming plantation in rows spaced 3m apart in all cases)</strong></td> <td> <p><em>Eg</em>: 900 trees/ha</p> <p><em>Pp</em>: 450 trees/ha</p> </td> <td> <p><em>Eg</em>: 1400 trees/ha</p> <p><em>Pp</em>: 1250 trees/ha</p> </td> </tr> <tr> <td> <p><strong>Average heights and corresponding ages of plantation trees</strong></p> </td> <td> <p><em>Eg</em>: 12m (6 years old)</p> <p><em>Pp</em>: 10m (15 years old)</p> </td> <td> <p><em>Eg</em>: 12m (6 years old)</p> <p><em>Pp</em>: 10m (15 years old)</p> </td> </tr> <tr> <td> <p><strong>Maximum heights and corresponding fully-matured ages of plantation trees</strong></p> </td> <td> <p><em>Eg</em>: 20m (11 years old)</p> <p><em>Pp</em>: 30m (40 years old)</p> </td> <td> <p><em>Eg</em>: 20m (11 years old)</p> <p><em>Pp</em>: 30m (40 years old)</p> </td> </tr> <tr> <td> <p><strong>Diameter at chest level (DCL – 1,3m) of plantation trees (average/maximum)</strong></p> </td> <td> <p><em>Eg</em>: 15cm/25cm</p> <p><em>Pp</em>: 20cm/50cm</p> </td> <td> <p><em>Eg</em>: 15cm/25cm</p> <p><em>Pp</em>: 20cm/50cm</p> </td> </tr> <tr> <td> <p><strong>Natural density of herbaceous plants</strong></p> </td> <td> <p>30% of woodland area</p> </td> <td> <p>30% of woodland area</p> </td> </tr> <tr> <td> <p><strong>Natural density of bush and shrubbery</strong></p> </td> <td> <p>30% of woodland area</p> </td> <td> <p>30% of woodland area</p> </td> </tr> <tr> <td> <p><strong>Natural density of arboreal plants (not part of plantation)</strong></p> </td> <td> <p>5% of woodland area</p> </td> <td> <p>5% of woodland area</p> </td> </tr> </tbody> </table> <ul> </ul>
PIANO-AUG - Piano Multipitch Estimation with Augmentation Dataset
<p>This dataset includes a 29 minutes long piano recording with a total of 4584 notes. The recording includes sections of single notes as well as groups of simultaneously sounding notes (intervals, chords), which ascend chromatically within the MIDI pitch range [21,92].</p> <p>In particular, the segments cover single notes, intervals (two-note groups), as well as common three- voiced and four-voiced chords. It must be noted that we do not investigate chord inversions here, all chords are used in their root position.</p> <p>The audio is created by rendering the MIDI file using the “Grand Piano” plugin in Ableton Live 8 with default settings except from the “Reverb Amount” being set to zero.</p> <p>The <a href="https://github.com/spotify/pedalboard">pedalboard</a> python library (0.4.1) can be used to create multiple augmented versions of the unprocessed piano recording. In particular, we create pairs of mild (-) and heavy (+) augmentations for each of the compression (comp), gain (gain), low-pass filter (lpf), and reverb (rev) effects using <a href="https://github.com/jakobabesser/piano_aug/blob/main/create_augmented_versions.py">this Python script</a></p> <p>Further details are provided at <a href="https://github.com/jakobabesser/piano_aug">https://github.com/jakobabesser/piano_aug</a>.</p>
data-augmentation-ner-datasets
<p>Dataset preparation and augmentation results produced in the context of evaluating data augmentation for Named Entity Recognition over the German legal domain.</p> <p>Detailed information can be found on the <a href="https://github.com/fusion-jena/data-augmentation-ner-legal">Github page</a>.</p> <p>This dataset is based on the <a href="http://github.com/elenanereiss/Legal-Entity-Recognition">German LER dataset</a> that is freely available under the <a href="https://creativecommons.org/licenses/by/4.0/deed.en">CC-BY 4.0 license</a></p>
Jacobaea vulgaris and meadow Augmented image classification dataset (binary)
<h3>General Information</h3> <p>Total instances: 117008<br>Instances in the Jacobaea vulgaris class: 58504 <br>Instances in the Meadow class: 58504<br>Image sizes from 224x224 pixels on three color channels (RGB)</p> <p><br>Performance increase training a ResNet50 on the base dataset versus the same architecture on the augmented data set shared here: +3,79 percent points in ROC AUC on an independent test set with 240 instances.<br><br></p> <h3>Data Generation and Source</h3> <p>The initial images in this dataset were taken as part of the project “UAV-basiertes Grünlandmonitoring auf Bestands- und Einzelpflanzenebene” (engl. “UAV-based Grassland Monitoring at Population and Individual Plant Level”), financed by the Authority for Economy, Transport, and Innovation of Hamburg. <br>In September 2018, flights with an octocopter were conducted over two extensively used grassland areas in the urban area of Hamburg.</p> <p>In my master's thesis at <a href="https://www.tu.berlin/dams">DAMS Lab</a> at TU Berlin, I evaluated the effect of different augmentation strategies for Jacobaea vulgaris image classification on the several performance metrics (most importantly the ROC AUC score). The identified augmentation strategies are -besides to performance based selection- also selected based on domain knowledge, which I acquired during the research for my master thesis. </p> <p>Additional information about the initial image generation process is to be found <a href="https://archiv.geomv.de/geoforum/2019/doc/Tagungsband_GeoForum-MV-2019_eBook.pdf">here </a> [p. 45–53] and <a href="https://www.auf.uni-rostock.de/storages/uni-rostock/Alle_AUF/AUF/GG/PDF/gruenlandmonitoring/2019-12-12-FHH-Workshop_Vortrag_Zacharias.pdf">here</a>. </p> <h3> </h3> <h3>Augmentations applied</h3> <ul> <li>Gaussian Noise: For the Gaussian noise augmentation, the mean of the added noise is set to zero. The lower and upper bounds for the random variance of the noise are 20.4663 and 54.0395 respectively. The bounds were identified by hyperparameter tuning. The search space for the lower bound was set from 5 to 30 and for the upper bound from 31 to 100. Those two search spaces were defined by visual inspection of the effects of applying Gaussian noise with different variance values to images of both classes. The Gaussian noise is sampled for each color channel individually. </li> <li>Random Brightness and Contrast: The brightness will randomly be increased or decreased by a factor ranging from 0.7010 to 1.2990. The The contrast will also be randomly increased by a factor ranging from 0.5775 to 1.4225. Those two ranges were identified using hyperparameter tuning. The search space for the maximal percentual increase or decrease of brightness and contrast was individually set from 1% to maximally 50% increase or decrease.</li> <li>Cutout Dropout: In this augmentation method a certain percentage of the input image is getting covered by black patches. The patches have a certain size in pixels, the implementation of this technique in this thesis uses square patches. The black patches are then randomly introduced into the image, by randomly alloacting the<br>patches across the image and then setting the corresponding pixel values to zero. The iamge is getting covered with patches until the cover percentage is reached. We<br>set percentage of the image to be randomly covered by black patches to 56.76%. The size of the patches, which randomly cover the image, is set to 4 pixels. A<br>good illustration of this is found in figure 4.2. The augmentation technique is inspired by the research proposed by Devries et al.[8]. Both values were identified by hyperparameter tuning. The search space for the patch size in pixels is categorical and includes the values [1, 2, 4, 7, 8, 14, 16, 28]. Those values all are multiples of 224, which is the image width and height in pixels. The patch size needs to be a multiple of the width and height in order to be suitable for the algorithm implementation. The search space for the cover percentage of the image had been set from 1% to 60%. This search space limits narrows the search down to a space where still a big part of the image is uncovered. The algorithm rearranges the image into a two dimensional grid and randomly masks rows of this grid by setting the pixel values in this row to zero. Then, the image gets rearranged, now with the randomly generated patches included.</li> <li>Random Saturation: The saturation of each pixel is randomly getting shifted. The upper bound for randomly shifting<span> </span>the saturation value of each pixel is set to 231.689%. This value was identified using hyperparameter tuning. An upper limit of the maximal saturation shift had been set to 40% shift in either direction for hyperparameter tuning.</li> <li>Horizontal Flip: The image gets flipped along the horizontal axis. </li> <li>Vertical Flip: The image gets flipped along the vertical axis.</li> <li>Random Rotation 90 degrees: Randomly rotates the image by a k-fold of 90 degrees, whereby k = {0, 1, 2, 3}.</li> </ul> <p> </p> <p>All augmentation methods and with their tuned augmentation hyperparameters (if existent) are applied to an image from the test set in figure 4.2. With the seven identified<br>augmentation techniques a dataset of 800% the size of the original dataset is created. The Augment model is trained on exactly this dataset. Of course next to the augmented images, the dataset still includes the original, unaugmented images. TensorFlow, along with additional libraries including Optuna for hyperparameter optimization and Albumentations for image augmentation, were used in for the implementation of this project.</p> <p> </p> <h3>Rational behind the augmentations applied</h3> <ul> <li>Random Rotation, Vertical and Horizontal Flip: These three augmentation strategies were chosen to make the classifier less sensitive to the orientation of the plant. The goal is to train a model that can classify plants regardless of their orientation. In order to achieve this effectively across different orientations, vertical flips, horizontal flips, and random 90-degree rotations are chosen for evaluation.</li> <li>Random Saturation: The varying saturation of the images simulates different levels of chlorophyll in the leaves, which is responsible for the green color of the<br>leaves and the intensity of this color. The color of the plant parts (leaves, stems, and flowers) is also influenced by factors such as soil, sun, weed density and pressure, location, and water availability. Varying the saturation of the images simulates changes in these factors.</li> <li>Gaussian Noise: By adding noise, in this case Gaussian noise, different lighting conditions are simulated when capturing the images. We specifically chose Gaussian<br>noise because it is common in many real-world scenarios and is based on the Central Limit Theorem, which states that the sum of many independent random variables.<br>tends to be normally distributed. This makes Gaussian noise a logical choice for simulating real-world random noise.</li> <li>Random Brightness Contrast: The Random Brightness and Random Contrast Augmentation uses brightness to mimic varying lighting conditions and contrast to highlight differences between plants by contrasting them more strongly, thereby highlighting their edges. This approach for highlighting edges is of course much more subtle than the canny edge detection augmentation. This augmentation method combines a weak focus on edges with variations in lighting conditions in one approach. The random contrast is a much softer approach for highlighting edges of plants, compared to the Canny edge detection augmentation. The other features in the images do not get changed that much, compared to the changes from edge detection augmentation.</li> <li>Cutout Dropout: The cutout augmentation simulates random occlusion by other plants. These occlusions are common and expected. Jacobaea vulgaris plants may be partially or completely obscured by other plants during image capturing. This augmentation technique makes the models more robust to random occlusion.</li> </ul> <h3> </h3> <h3>Data License</h3> <p>The dataset is licensed under the license CC BY 4.0. The attributor of the data is the Chair of Geodesy and Geoinformatics at the University of Rostock. The data was created within the scope of the project 'UAV-based Grassland Monitoring at Population and Individual Plant Level', financed by the Authority for Economy, Transport, and Innovation of Hamburg.</p>
Augmented dataset of rumours and non-rumours for rumour detection
<p> </p> <p>This data set contains a collection of Twitter rumours and non-rumours during six real-world events: 1) 2013 Boston marathon bombings, 2) 2014 Ottawa shooting, 3) 2014 Sydney siege, 4) 2015 Charlie Hebdo Attack, 5) 2014 Ferguson unrest, and 6) 2015 Germanwings plane crash</p> <p>The data set is an augmented data set of the PHEME dataset of rumours and non-rumours based on two data sets: the PHEME data [2] (downloaded via <a href="https://figshare.com/articles/PHEME_dataset_for_Rumour_Detection_and_Veracity_Classification/6392078">https://figshare.com/articles/PHEME_dataset_for_Rumour_Detection_and_Veracity_Classification/6392078</a>), and the CrisisLexT26 data [3] (downloaded via <a href="https://github.com/sajao/CrisisLex/tree/master/data/CrisisLexT26/2013_Boston_bombings">https://github.com/sajao/CrisisLex/tree/master/data/CrisisLexT26/2013_Boston_bombings</a>).</p> <p> </p> <p><strong>PHEME-Aug v2.0 (aug-rnr-data_filtered.tar.bz2 and aur-rnr-data_full.tar.bz2) </strong>contains augmented data for all six events.</p> <p><strong>aug-rnr-data_full.tar.bz2</strong> contains source tweets and replies without temporal filtering. Please refer to [1] for details about temporal filtering. The statistics are as follows:</p> <p>* 2013 Boston marathon bombings: 392 rumours and 784 non-rumours </p> <p>* 2014 Ottawa shooting: 1,047 rumours and 2,072 non-rumours </p> <p>* 2014 Sydney siege: 1,764 rumours and 3,530 non-rumours </p> <p>* 2015 Charlie Hebdo Attack: 1,225 rumours and 2,450 non-rumours </p> <p>* 2014 Ferguson unrest: 737 rumours and 1,476 non-rumours </p> <p>* 2015 Germanwings plane crash: 502 rumours and 604 non-rumours </p> <p> </p> <p><strong>aug-rnr-data_filtered.tar.bz2 </strong>contains source tweets, replies, and retweets after temporal filtering and deduplication. Please refer to [1] for details. The statistics are as follows:</p> <p>* 2013 Boston marathon bombings: 323 rumours and 645 non-rumours </p> <p>* 2014 Ottawa shooting: 713 rumours and 1,420 non-rumours </p> <p>* 2014 Sydney siege: 1,134 rumours and 2,262 non-rumours </p> <p>* 2015 Charlie Hebdo Attack: 812 rumours and 1,673 non-rumours </p> <p>* 2014 Ferguson unrest: 471 rumours and 949 non-rumours </p> <p>* 2015 Germanwings plane crash: 375 rumours and 402 non-rumours </p> <p> </p> <p>The data structure follows the format of the PHEME data [2]. Each event has a directory, with two subfolders, rumours and non-rumours. These two folders have folders named with a tweet ID. The tweet itself can be found on the 'source-tweet' directory of the tweet in question, and the directory 'reactions' has the set of tweets responding to that source tweet. Also each folder contains ‘aug_complete.csv’ and ‘reference.csv'.</p> <p>'aug_complete.csv' file contains the metadata (tweet ID, tweet text, timestamp, and rumour label) of augmented tweets before deduplication and filtering tweets without context (i.e., replies).</p> <p>'reference.csv' file contains manually annotated reference tweets [2, 3].</p> <p> </p> <p><strong>If you use our augmented data (PHEME-Aug v2.0), please also cite:</strong></p> <p>[1] Han S., Gao, J., Ciravegna, F. (2019). "Neural Language Model Based Training Data Augmentation for Weakly Supervised Early Rumor Detection", The 2019 IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining (ASONAM 2019), Vancouver, Canada, 27-30 August, 2019</p> <p>==============================================================================================</p> <p>[2] Kochkina, E., Liakata, M., & Zubiaga, A. (2018). All-in-one: Multi-task Learning for Rumour Verification. COLING.</p> <p>[3] Olteanu, A., Vieweg, S., & Castillo, C. (2015, February). What to expect when the unexpected happens: Social media communications across crises. In <em>Proceedings of the 18th ACM conference on computer supported cooperative work & social computing</em> (pp. 994-1009). ACM</p>
Dataset Interactive Audio Augmented Reality in Participatory Performance _Please Confirm You Are Not A Robot_
<p>This dataset gathers the different data from the study "Interactive Audio Augmented Reality in Participatory Performance". </p>
Dataset for manuscript 'CeyeHao: AI-driven microfluidic flow programming with hierarchically assembled obstacles in microchannel and receptive-field-augmented neural network'
<p>This dataset contains:<br>1. The dataset used to train the models related to the manuscript 'CeyeHao: AI-driven microfluidic flow programming using hierarchically assembled obstacles in microchannel with receptive-field-augmented neural network'.<br>2. A checkpoint of trained 'CEyeNet' proposed in the manuscript.<br>3. Example microchannels designed in the manuscript to produce semantic flow profiles</p> <p>This dataset is intended for research and academic purpose.</p> <p>Detailed description please refer to the enclosed ReadMe.txt.</p>
Data augmentation for Multi-Classification of Non-Functional Requirements - Dataset
<p>There are four datasets:</p> <p>1.Dataset_structure indicates the structure of the datasets, such as column name, type, and value.</p> <p>2. Spanish_promise_exp_nfr_train and Spanish_promise_exp_nfr_test are the non-functional requirements of the Promise_exp[1] dataset translated into the Spanish language.</p> <p>3. Balanced_promise_exp_nfr_train is the new balanced dataset of Spanish_promise_exp_nfr_train, in which the Data Augmentation technique with chatGPT was applied to increase the requirements with little data and random undersampling was used to eliminate requirements.</p> <p>The labeling schema, similar to PROMISE NFR, includes the following categories: A: Availability, PO: Portability, L: Legal, FT: Fault tolerance, SC: Scalability, MN: Maintainability, LF: Look and feel, PE: Performance, O: Operational. US: Usability, and SE: Security.</p>
Paired Fusion Augmented Dataset for Vehicle Extraction and Counting (Domino Dataset)
<p>This dataset try to expedite the deep learning researcher's task of a model training to extract vehicles from aerial images in an urban environment. Vehicles included in the dataset are motorcycles and cars of any type, number of wheels and color. The specific process of acquisition, enhancing, fusion and augmentation is presented. Inclusion of height of cars using a Digital Surface Model (DSM) is described and comparison of the application of a U-net segmentation model over non height and height dataset is shown.</p>
Variable Misuse tool: Dataset for data augmentation (3)
<p>Dataset used for data augmentation in the training phase of the Variable Misuse tool. It contains some source code files extracted from third-party repositories.</p>
Variable Misuse tool: Dataset for data augmentation (7)
<p>Dataset used for data augmentation in the training phase of the Variable Misuse tool. It contains some source code files extracted from third-party repositories.</p>
Variable Misuse tool: Dataset for data augmentation (2)
<p>Dataset used for data augmentation in the training phase of the Variable Misuse tool. It contains some source code files extracted from third-party repositories.</p>
ForTrunkDetV2 - Image dataset of visible and thermal annotated images for forest tree trunk detection (augmented version)
<p>This dataset is an augmented version of an existing dataset (<a href="https://doi.org/10.5281/zenodo.5213824">10.5281/zenodo.5213824</a>), composed by visible and thermal images with trunk annotations. The images were acquired in three different portuguese forests and were captured by five different cameras:</p> <ul> <li>GoPro Hero6</li> <li>Allied Mako G-125</li> <li>FLIR M232</li> <li>ZED Stereo</li> <li>OAK-D</li> </ul> <p>The augmented images and their annotations are stored in an archive with the following structure:</p> <p><em>main_directory/</em></p> <ol> <li><em>annotations/</em> <ol> <li><em>pascalvoc/</em> <ul> <li>PascalVOC annotation files</li> <li>...</li> </ul> </li> <li><em>yolo/</em> <ul> <li>YOLO annotation files</li> <li>...</li> </ul> </li> </ol> </li> <li><em>images/</em> <ul> <li>Image files</li> <li>...</li> </ul> </li> </ol> <p><br> Each annotation file links to its image by the file name, so if an image is named <strong>"img12345.jpg</strong><em><strong>"</strong></em>, its annotation files are named as <strong>"img12345.xml"</strong> (for PascalVOC format) and <strong>"img12345.txt"</strong> (for YOLO format).<br> <br> The dataset contains original and augmented images. The original images' names follow the pattern <strong>"img_******.jpg"</strong>, where in the place of the asterisks are numbers. The remaining images are the augmented ones.</p> <p>Also, the subsets that were used to train, validate and test some deep learning models are available in three .TXT files (train.txt, val.txt and test.txt), where each file line corresponds to an image name.</p>
A-MAPS: Augmented MAPS Dataset with Rhythm and Key Annotations
<p>The MAPS dataset is one of the most used benchmark dataset for automatic music transcription. We propose here an updated version of the ground truth MIDI files, containing, on top of the original pitch, onset and offsets, additional annotations.</p> <p>The annotations include:</p> <ul> <li> <p>Tempo curve</p> </li> <li> <p>Time signature</p> </li> <li> <p>Durations of notes in fraction of a quarter note (some of them are approximate)</p> </li> <li> <p>Key signature (always written as the major relative)</p> </li> <li> <p>Separate left and right hand staff</p> </li> <li> <p>Text annotations from the score (tempo indications, coda...).</p> </li> </ul> <p>If you use these annotations in a published research project, please cite:<br> Adrien Ycart and Emmanouil Benetos. “A-MAPS: Augmented MAPS Dataset with Rhythm and Key Annotations” <em>19th International Society for Music Information Retrieval Conference Late Breaking and Demo Papers, </em>September 2018, Paris, France.</p> <p>More information is available at: <a href="http://c4dm.eecs.qmul.ac.uk/ycart/a-maps.html">http://c4dm.eecs.qmul.ac.uk/ycart/a-maps.html</a></p> <p><strong>Version 1.2 (2019-03-11):</strong></p> <ul> <li>Add 3 missing MAPS files to the dataset: MAPS_MUS-chpn-e01_ENSTDkCl.mid, MAPS_MUS-chpn-e01_SptkBGAm.mid, MAPS_MUS-chpn-e01_StbgTGd2.mid. <strong>WARNING:</strong> <strong>They correspond to the same piece as MAPS_MUS-chpn_op10_e01_AkPnCGdD.mid</strong></li> <li>Tempo curve slightly edited to avoid unnatural tempo changes (see <a href="http://c4dm.eecs.qmul.ac.uk/ycart/a-maps.html">this page</a> for more info).</li> </ul> <p><strong>Version 1.1 (2018-09-19):</strong></p> <ul> <li>BUG FIXED: in some files, the time signature, key signature and text annotations used to be incorrectly shifted by one quarter note. This is now fixed.</li> <li>Sustain pedal activations are now the same as in the MAPS dataset, they are all in the "Piano left" track.</li> </ul> <p> </p>
DATASET Volumetric Stability of Biphasic HAp/b-TCP/Collagen Implants in Malar Zone Augmentation: A CBCT Study in Orthognathic Surgery Patients
<p>This dataset repository contains data and analysis scripts related to the study on the volumetric stability of biphasic HAp/b-TCP/collagen implants used for malar zone augmentation in orthognathic surgery patients. The study evaluates the changes in implant volume over time using Cone Beam Computed Tomography (CBCT) scans. The dataset includes CSV files with clinical data and HTML files documenting the analysis procedures.</p> <h3> </h3>
Dataset for a flume experiment on sediment augmentation
<p>Dataset for the paper "Influence of hydrograph shape and sediment augmentation repetition frequency on sediment transport dynamics and bed morphology evolution". It includes coordinates of the defined areas of interest and digital elevation models from the scanned channel topographies.</p>
Comparing 2D and Augmented Reality Visualizations for Microservice System Understandability: Protocol And Dataset
<p>Comparing 2D and Augmented Reality Visualizations for Microservice System Understandability: Dataset.</p> <p>This dataset includes:</p> <ol> <li>The tools executable files and JSON representations.</li> <li>The tools screenshots.</li> <li>The participants classification data.</li> <li>Questionnaire Form.</li> <li>Training Materials.</li> <li>Testing Tasks.</li> </ol> <p> </p> <p><strong>Our Paper is at ICPC conference with title</strong>: Comparing 2D and Augmented Reality Visualizations for Microservice System Understandability: A Controlled Experiment</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.