Papers with Code 아카이브

Classification

태스크의 하위 분류와 논문·코드·평가·기법·데이터셋을 탐색하세요.

하위 태스크
24
평가표
72
보존된 평가 기록
134

Classification is the task of categorizing a set of data into predefined classes or groups. The aim of classification is to train a model to correctly predict the class or group of new, unseen data. The model is trained on a labeled dataset where each instance is assigned a class label. The learning algorithm then builds a mapping between the features of the data and the class labels. This mapping is then used to predict the class label of new, unseen data points. The quality of the prediction is usually evaluated using metrics such as accuracy, precision, and recall.

데이터셋

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

DEEP-VOICE: DeepFake Voice Recognition

DEEP-VOICE: Real-time Detection of AI-Generated Speech for DeepFake Voice Conversion This dataset contains examples of real human speech, and DeepFake versions of those speeches by using Retrieval-based Voice Conversion. Can machine learning be used to detect when speech is AI-generated? ## Introduction There are growing implications surrounding generative AI in the speech domain that enable voice cloning and real-time voice conversion from one individual to another. This technology poses a significant ethical threat and could lead to breaches of privacy and misrepresentation, thus there is an urgent need for real-time detection of AI-generated speech for DeepFake Voice Conversion. To address the above emerging issues, we are introducing the DEEP-VOICE dataset. DEEP-VOICE is comprised of real human speech from eight well-known figures and their speech converted to one another using Retrieval-based Voice Conversion. For each speech, the accompaniment ("background noise") was removed before conversion using RVC. The original accompaniment is then added back to the DeepFake speech: Overview of the Retrieval-based Voice Conversion process to generate DeepFake speech with Ryan Gosling's speech converted to Margot Robbie. Conversion is run on the extracted vocals before being layered on the original background ambience. (Above: Overview of the Retrieval-based Voice Conversion process to generate DeepFake speech with Ryan Gosling's speech converted to Margot Robbie. Conversion is run on the extracted vocals before being layered on the original background ambience.) ## Dataset There are two forms to the dataset that are made available. First, the raw audio can be found in the "AUDIO" directory. They are arranged within "REAL" and "FAKE" class directories. The audio filenames note which speakers provided the real speech, and which voices they were converted to. For example "Obama-to-Biden" denotes that Barack Obama's speech has been converted to Joe Biden's voice. Second, the extracted features can be found in the "DATASET-balanced.csv" file. This is the data that was used in the below study. The dataset has each feature extracted from one-second windows of audio and are balanced through random sampling. **Note: ** All experimental data is found within the "KAGGLE" directory. The "DEMONSTRATION" directory is used for playing cropped and compressed demos in notebooks due to Kaggle's limitations on file size. A potential use of a successful system could be used for the following: Usage of the real-time system. The end user is notified when the machine learning model has processed the speech audio (e.g. a phone or conference call) and predicted that audio chunks contain AI-generated speech. (Above: Usage of the real-time system. The end user is notified when the machine learning model has processed the speech audio (e.g. a phone or conference call) and predicted that audio chunks contain AI-generated speech.) ## Kaggle The dataset is available on the Kaggle data science platform. The Kaggle page can be found by clicking here: Dataset on Kaggle ## Attribution This dataset was produced from the study "Real-time Detection of AI-Generated Speech for DeepFake Voice Conversion" The preprint can be found on ArXiv by clicking here: Real-time Detection of AI-Generated Speech for DeepFake Voice Conversion ## License This dataset is provided under the MIT License: Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

DeepGraviLens

DeepGraviLens is a data set of simulated gravitational lenses consisting of images associated with brightness variation time series. In this dataset, both non-transient and transient phenomena (supernovae explosions) are simulated.

DeepParliament

DeepParliament is a legal domain Benchmark Dataset that gathers bill documents and metadata and performs various bill status classification tasks. The dataset text covers a broad range of bills from 1986 to the present and contains richer information on parliament bill content. There are a total of 5329 documents where 4223 are in the train and 1106 are in the test dataset. Each bill document contains many sentences in both cases, and the document’s length varies greatly. Source: DeepParliament: A Legal domain Benchmark & Dataset for Parliament Bills Prediction Image Source: https://arxiv.org/pdf/2211.15424v1.pdf

DEJAN Synthetic Taxonomy Dataset

This dataset contains synthetic text data generated to train models for text generation. The data is created using a large language model (LLM), specifically the GEMMA-2-9B-IT model, and is designed to encompass a diverse range of scenarios and writing styles.

Dhoroni

Climate change poses critical challenges globally, disproportionately affecting low-income countries that often lack resources and linguistic representation on the international stage. Despite Bangladesh's status as one of the most vulnerable nations to climate impacts, research gaps persist in Bengali-language studies related to climate change and NLP. To address this disparity, we introduce ধরণী (Dhoroni), a novel Bengali (Bangla) climate change and environmental news dataset, comprising a 2300 annotated Bangla news articles, offering multiple perspectives such as political influence, scientific/statistical data, authenticity, stance detection, and stakeholder involvement. Furthermore, we present an in-depth exploratory analysis of Dhoroni and introduce BanglaBERT-Dhoroni family, a novel baseline family for climate stance detection in Bangla, fine-tuned on our dataset. This research contributes significantly to enhancing accessibility and analysis of climate discourse in Bengali (Bangla), addressing crucial communication and research gaps in climate-impacted regions like Bangladesh with 180 million people.

DIGITal

Digitally Generated Numerals (DIGITal) ## Description The Digitally Generated Numerals (DIGITal) dataset consists of 100,000 image pairs representing digits from 0 to 9. These image pairs include both low and high-quality versions, with a resolution of 128x128 pixels. ## Languages The dataset is generated from fonts in the following languages: - Arabic - Chinese (Simplified) - Farsi and Urdu - Gurmukhi - Gujarati - Tibetan - ARDIS (Sweden) - ISI Bangla - Bangla Lekha - Kannada - English ## Image Resolution - resolution: 128x128 pixels ## Image Pairs - Each digit (0 to 9) has 1000 associated images. ## Total Images - Total images: 100,000 ## Samples | !Image 1 | !Image 2 | !Image 3 | !Image 4 | !Image 5 | |:-----------------------:|:-----------------------:| | Arabic | ARDIS | Chinese | Farsi & Urdu | Gujarati | | !Image 5 | !Image 6 | !Image 7 | !Image 8 | !Image 5 | |:-----------------------:|:-----------------------:| | ISI Bangla & BanglaLekha | Kannada | Tibetan | English | Gurmukhi This dataset is a valuable resource for tasks related to digit recognition, font analysis, and super-resolution. Researchers and practitioners can utilize it for various applications in computer vision and machine learning.

DiscoEval

Dataset Summary The DiscoEval is an English-language Benchmark that contains a test suite of 7 tasks to evaluate whether sentence representations include semantic information relevant to discourse processing. The benchmark datasets offer a collection of tasks designed to evaluate natural language understanding models in the context of discourse analysis and coherence. Dataset Sources Arxiv: A repository of scientific papers and research articles. Wikipedia: An extensive online encyclopedia with articles on diverse topics. Rocstory: A dataset consisting of fictional stories. Ubuntu IRC channel: Conversational data extracted from the Ubuntu Internet Relay Chat (IRC) channel. PeerRead: A dataset of scientific papers frequently used for discourse-related tasks. RST Discourse Treebank: A dataset annotated with Rhetorical Structure Theory (RST) discourse relations. Penn Discourse Treebank: Another dataset with annotated discourse relations, facilitating the study of discourse structure

E2E Refined

E2E Refined is a dataset for sentence classification. It consists of 40,560 examples for training, 4,489 for validation, and 4,555 for test. It is a refined version of the well-known MR-to-text E2E dataset where many deletion/insertion/substitution errors has been fixed.

ENSeg

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

Eth-ICO

The sampled 2-hop subgraphs centered on ICO-wallet accounts on the Ethereum Interaction graph.

Eth-Mining

The sampled 2-hop subgraphs centered on Mining accounts on the Ethereum Interaction graph.

EuroSAT-C

EuroSAT-C is an open-source data set comprising algorithmically generated corruptions applied to the EuroSAT test set following the concept of ImageNet-C. It comprises 19 different corruptions (15 test corruptions and 4 validation corruptions) spanning 5 severity levels resulting in 108,000 images for the validation set and 405,000 images for the test set. For further information on the corruptions visit the original GitHub repository of ImageNet-C.

FairFD

A Racial Fairness Benchmark Dataset for Face Forgery Detection.

FedNLP

We collect the various forms of Federal Reserve communications. - fomc_doc: FOMC post-meeting statements, Minutes, and press conferences. (122 docs) - speaker_doc: FOMC members’ speeches from over 30 different websites. (1300 docs) - period: Jan 2015 - Jul 2020 Open In Colab

FinSen

Enhancing Financial Market Predictions: Causality-Driven Feature Selection This paper introduces FinSen dataset that revolutionizes financial market analysis by integrating economic and financial news articles from 197 countries with stock market data. The dataset’s extensive coverage spans 15 years from 2007 to 2023 with temporal information, offering a rich, global perspective 160,000 records on financial market news. Our study leverages causally validated sentiment scores and LSTM models to enhance market forecast accuracy and reliability. # Our FinSen Dataset arXiv Pytorch 1.5 License: MIT This repository contains the dataset for Enhancing Financial Market Predictions: Causality-Driven Feature Selection, which has been accepted in ADMA 2024. If the dataset or the paper has been useful in your research, please add a citation to our work: @article{liang2024enhancing, title={Enhancing Financial Market Predictions: Causality-Driven Feature Selection}, author={Liang, Wenhao and Li, Zhengyang and Chen, Weitong}, journal={arXiv e-prints}, pages={arXiv--2408}, year={2024} } ### Datasets [FinSen] can be downloaded manually from the repository as csv file. Sentiment and its score are generated by FinBert model from the Hugging Face Transformers library under the identifier "ProsusAI/finbert". (Araci, Dogu. "Finbert: Financial sentiment analysis with pre-trained language models." arXiv preprint arXiv:1908.10063 (2019).) We only provide US for research purpose usage, please contact w.liang@adelaide.edu.au for other countries (total 197 included) if necessary. We also provide other NLP datasets for text classification tasks here, please cite them correspondingly once you used them in your research if any. 1. 20Newsgroups. Joachims, T., et al.: A probabilistic analysis of the rocchio algorithm with tfidf for text categorization. In: ICML. vol. 97, pp. 143–151. Citeseer (1997) 2. AG News. Zhang, X., Zhao, J., LeCun, Y.: Character-level convolutional networks for text classification. Advances in neural information processing systems 28 (2015) 3. Financial PhraseBank. Malo, P., Sinha, A., Korhonen, P., Wallenius, J., Takala, P.: Good debt or bad debt: Detecting semantic orientations in economic texts. Journal of the Association for Information Science and Technology 65(4), 782–796 (2014) ### Dataloader for FinSen We provide the preprocessing file finsen.py for our FinSen dataset under dataloaders directory for more convienient usage. ### Models - Text Classification 1. DAN-3. 2. Gobal Pooling CNN. ### Models - Regression Prediction 1. LSTM ### Using Sentiment Score from FinSen Predict Result on S&P500 ### Dependencies The code is based on PyTorch under code frame of https://github.com/torrvision/focal_calibration, please cite their work if you found it is useful. :smiley: ☺ Happy Research !

FMC-MWO2KG

The Failure Mode Classification dataset released in the paper "MWO2KG and Echidna: Constructing and exploring knowledge graphs from maintenance data" by Stewart et al. The goal is to label a given observation (made by a maintainer) with the corresponding Failure Mode Code. Each row contains an observation made by a maintainer, followed by a comma, followed by the Failure Mode, for example: falure,Breakdown As they are written in technical language, there are often spelling/grammatical/tokenisation errors made in the observations - these are typical of maintenance work orders. The dataset comprises 502 (observation, label) pairs (for training), 62 pairs (for validation) and 62 pairs (for testing). The labels are taken from a set of 22 failure mode codes from ISO 14224. In order to pull a list of observations in which to label, we ran MWO2KG over the data once and exported a list of all entities labelled as ‘observation’ (such as ‘leaking’, ‘not working’) by the Named Entity Recognition model. We then removed all results that were incorrectly predicted as observations by the NER model and proceeded to label each observation with the most appropriate failure mode code using a text editor. The source code of the above paper (which also includes this dataset) is located on GitHub. The direct link to the data (train.txt, dev.txt, and test.txt) is available here.

Food Recall Incidents Dataset

The Food Recall Incidents dataset consists of 7,546 short texts (from 5 to 360 characters each), which are the titles of food recall announcements (therefore referred to as title), crawled from 24 public food safety authority websites by Agroknow. The texts are written in 6 languages, with English (6,644) and German (888) being the most common, followed by French (8), Greek (4), Italian (1) and Danish (1). Most of the texts have been authored after 2010 and they describe recalls of specific food products due to specific hazards. Experts manually classified each text to four groups of classes describing hazards and products on two levels of granularity: - hazard: fine-grained description of the hazards mentioned in the texts comprising 261 classes; - hazard-category: categorized version of the hazard classification task comprising 10 classes; - product: fine-grained description of the products mentioned in the texts comprising 1,256 classes; - product-category: categorized version of the product classification task comprising 22 classes. The columns hazard-title and product-title comprise character spans, generated based on feature importance of a Logistic Regression (LR) classifier. These signify parts of the title that are important for hazard and product classification. Due to their very low support for many classes, the fine-grained tasks of hazard and product classification may require further pre-processing (e.g. label clustering or filtering), dependent on the application. The dataset comprises also metadata, such as the release date of the text (year, month, day), the language of the text (language), and the country of issue (country).

Forex News Annotated Dataset for Sentiment Analysis

This dataset contains news headlines relevant to key forex pairs: AUDUSD, EURCHF, EURUSD, GBPUSD, and USDJPY. The data was extracted from reputable platforms Forex Live and FXstreet over a period of 86 days, from January to May 2023. The dataset comprises 2,291 unique news headlines. Each headline includes an associated forex pair, timestamp, source, author, URL, and the corresponding article text. Data was collected using web scraping techniques executed via a custom service on a virtual machine. This service periodically retrieves the latest news for a specified forex pair (ticker) from each platform, parsing all available information. The collected data is then processed to extract details such as the article's timestamp, author, and URL. The URL is further used to retrieve the full text of each article. This data acquisition process repeats approximately every 15 minutes. To ensure the reliability of the dataset, we manually annotated each headline for sentiment. Instead of solely focusing on the textual content, we ascertained sentiment based on the potential short-term impact of the headline on its corresponding forex pair. This method recognizes the currency market's acute sensitivity to economic news, which significantly influences many trading strategies. As such, this dataset could serve as an invaluable resource for fine-tuning sentiment analysis models in the financial realm. We used three categories for annotation: 'positive', 'negative', and 'neutral', which correspond to bullish, bearish, and hold sentiments, respectively, for the forex pair linked to each headline. The following Table provides examples of annotated headlines along with brief explanations of the assigned sentiment. Examples of Annotated Headlines Forex Pair Headline Sentiment Explanation GBPUSD Diminishing bets for a move to 12400 Neutral Lack of strong sentiment in either direction GBPUSD No reasons to dislike Cable in the very near term as long as the Dollar momentum remains soft Positive Positive sentiment towards GBPUSD (Cable) in the near term GBPUSD When are the UK jobs and how could they affect GBPUSD Neutral Poses a question and does not express a clear sentiment JPYUSD Appropriate to continue monetary easing to achieve 2% inflation target with wage growth Positive Monetary easing from Bank of Japan (BoJ) could lead to a weaker JPY in the short term due to increased money supply USDJPY Dollar rebounds despite US data. Yen gains amid lower yields Neutral Since both the USD and JPY are gaining, the effects on the USDJPY forex pair might offset each other USDJPY USDJPY to reach 124 by Q4 as the likelihood of a BoJ policy shift should accelerate Yen gains Negative USDJPY is expected to reach a lower value, with the USD losing value against the JPY AUDUSD RBA Governor Lowe’s Testimony High inflation is damaging and corrosive Positive Reserve Bank of Australia (RBA) expresses concerns about inflation. Typically, central banks combat high inflation with higher interest rates, which could strengthen AUD. Moreover, the dataset includes two columns with the predicted sentiment class and score as predicted by the FinBERT model. Specifically, the FinBERT model outputs a set of probabilities for each sentiment class (positive, negative, and neutral), representing the model's confidence in associating the input headline with each sentiment category. These probabilities are used to determine the predicted class and a sentiment score for each headline. The sentiment score is computed by subtracting the negative class probability from the positive one.

FTR-18

FTR-18 is a multilingual rumour dataset on football transfer news. Transfer rumours are continuously published by sports media. They can both harm the image of player or a club or increase the player's market value. The proposed dataset includes transfer articles written in English, Spanish and Portuguese. It also comprises Twitter reactions related to the transfer rumours. FTR-18 is suited for rumour classification tasks and allows the research on the linguistic patterns used in sports journalism.

Gambling Address Dataset

Gambling Address Dataset is a collection of 10,423 gambling addresses that have transactions with gambling contracts. Moreover, 51,004 non-gambling addresses are also selected (such as exchanges, wallet addresses, etc.), making the gambling address dataset more complete. In the dataset, accounts are used to refer to addresses (e.g. 0xd1ce...edec95), where 1, 0, and -1 represent the gamble, non-gamble, and other types, respectively. Source: Who is Gambling? Finding Cryptocurrency Gamblers Using Multi-modal Retrieval Methods Image Source: https://arxiv.org/pdf/2211.14779v1.pdf

Gambling Contract Dataset

Gambling Contract Dataset is a collection of 260 gambling smart contracts from decentralized gambling websites, such as Dicether, Degens. At the same time, in order to construct the negative samples required for training, 1040 smart contracts that are not involved in gambling (e.g., erc20, erc721, mixer, etc.) are selected . In the dataset, accounts are used to refer to contracts (e.g. 0x3fe2b...f8a33f), where 1, 0, and -1 to represent the gamble, non-gamble, and other types, respectively. Source: Who is Gambling? Finding Cryptocurrency Gamblers Using Multi-modal Retrieval Methods Image Source: https://arxiv.org/pdf/2211.14779v1.pdf

GANGen-Detection

This dataset was created to test whether it's possible to build a general-purpose detector that can tell real images apart from fake ones generated by convolutional neural networks (CNNs), no matter which model or dataset was used to create the fake images. To do this, the authors collected fake images generated by 11 different CNN-based image generation models. These models represent a wide range of current image synthesis techniques and include: ProGAN StyleGAN BigGAN CycleGAN StarGAN GauGAN DeepFakes Cascaded Refinement Networks (CRN) Implicit Maximum Likelihood Estimation (IMLE) Second-order Attention Super-Resolution (SOAT-SR) Seeing-in-the-Dark (SID) The dataset includes fake images from each of these models and a set of real images, allowing for binary classification (real vs. fake). The study found that a standard image classifier (like a convolutional neural network) trained on fake images from just one generator (ProGAN) was able to detect fake images from other, completely different generators with surprising accuracy. This suggests that many CNN-generated images, even from different architectures, share common flaws that can be learned and detected. The dataset is useful for research in detecting synthetic media, improving image forensics, and understanding the weaknesses in current generative models. Code and pre-trained models were made available by the authors (https://github.com/chuangchuangtan/GANGen-Detection).

Gaze-CIFAR-10

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

Genocide Transcript Corpus (GTC): Topic-Based Paragraph Classification in Genocide-Related Court Transcripts

The Topic-Based Paragraph Classification in Genocide-Related Court Transcripts (GTC) dataset is the first reference corpus annotated with samples from genocide tribunals in different international criminal courts. It is made up of witness statements about violence experienced. The material consists of 1475 text passages with about 40 to 120 pages per transcript, covering 3 tribunals: the Extraordinary Chambers in the Courts of Cambodia (ECCC) - 438 pages, the International Criminal Tribunal for Rwanda (ICTR) - 566 pages, and the International Criminal Tribunal of the Former Yugoslavia (ICTY) - 416 pages. As no datasets of any kind containing genocide court transcripts have been published nor other forms of pre-structured or annotated text data in this field of research exist, the aim was to address this gap by providing a systematically annotated dataset. Potential use cases include genocide-related inquiry conducted by those who need better to access, explore, and search through extensive documentation on these topics including researchers, lawyers and other practitioners. Broadly, its stated aim is to serve 3 purposes: * (1) to provide a first reference corpus for the community * (2) to establish benchmark performances (using state-of-the-art transformer-based approaches) for the new classification task of paragraph identification of violence-related witness statements * (3) to explore first steps towards transfer learning within the domain

GLAMI-1M

We introduce GLAMI-1M: the largest multilingual image-text classification dataset and benchmark. The dataset contains images of fashion products with item descriptions, each in 1 of 13 languages. Categorization into 191 classes has high-quality annotations: all 100k images in the test set and 75% of the 1M training set were human-labeled. The paper presents baselines for image-text classification showing that the dataset presents a challenging fine-grained classification problem: The best scoring EmbraceNet model using both visual and textual features achieves 69.7% accuracy. Experiments with a modified Imagen model show the dataset is also suitable for image generation conditioned on text.

Google Ranked URLs Dataset

This dataset was curated for Search Engine Optimization (SEO) analysis tasks, including categorization and spam detection. It covers 12 diverse topics: basketball, books, cats, gardening, history, movies, music, recipes, sports, technology, travel, and weather. Some topics have hierarchical relationships, such as sports and basketball, while others are closely related (e.g., movies and music) or unrelated (e.g., basketball and gardening), with varying degrees of overlap among them. For each topic, approximately 300 search queries were generated using large language models (LLMs) like GPT, Llama, and Claude. The top 10 URLs from the Google Search Console’s search engine results page (SERP) were retrieved for each query.

Graph dataset MCF-7

Dataset introduced by Xifeng Yan et al. in: SIGMOD '08: Proceedings of the 2008 ACM SIGMOD international conference on Management of data, June 2008, Pages 433–444, https://doi.org/10.1145/1376616.1376662 The dataset is now hosted by TUD. The dataset consists of small molecules activities against breast cancer tumors.

Graph dataset MOLT-4

Dataset introduced by Xifeng Yan et al. in: SIGMOD '08: Proceedings of the 2008 ACM SIGMOD international conference on Management of data, June 2008, Pages 433–444, https://doi.org/10.1145/1376616.1376662 The dataset is now hosted by TUD. The dataset consists of small molecules activities against Leukemia tumors.

HOWS

HOWS-CL-25 (Household Objects Within Simulation dataset for Continual Learning) is a synthetic dataset especially designed for object classification on mobile robots operating in a changing environment (like a household), where it is important to learn new, never seen objects on the fly. This dataset can also be used for other learning use-cases, like instance segmentation or depth estimation. Or where household objects or continual learning are of interest. Our dataset contains 150,795 unique synthetic images using 25 different household categories with 925 3D models in total. For each of those categories, we generated about 6000 RGB images. In addition, we also provide a corresponding depth, segmentation, and normal image. The dataset was created with BlenderProc [Denninger et al. (2019)], a procedural pipeline to generate images for deep learning. This tool created a virtual room with randomly textured floors, walls, and a light source with randomly chosen light intensity and color. After that, a 3D model is placed in the resulting room. This object gets customized by randomly assigning materials, including different textures, to achieve a diverse dataset. Moreover, each object might be deformed with a random displacement texture. We use 774 3D models from the ShapeNet dataset [A. X. Chang et al. (2015)] and the other models from various internet sites. Please note that we had to manually fix and filter most of the models with Blender before using them in the pipeline! For continual learning (CL), we provide two different loading schemes: - Five sequences with five categories each - Twelve sequences with three categories in the first and two in the other sequences. In addition to the RGB, depth, segmentation, and normal images, we also provide the calculated features of the RGB images (by ResNet50) as used in our RECALL paper. In those two loading schemes, ten percent of the images are used for validation, where we ensure that an object instance is either in the training or the validation set, not in both. This avoids learning to recognize certain instances by heart. We recommend using those loading schemes to compare your approach with others. For further information and code examples, please have a look at our website: https://github.com/DLR-RM/RECALL.

HRPlanesV2

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

Human-ChatGPT texts

A dataset including texts by humans (labeled 0) and then rephrased by ChatGPT (labeled 1), created to train models for machine-generated text detection. It is a robust dataset - includes text of various lengths, and the human texts are taken from multiple sources.

iLur News Texts

iLur News Texts is a dataset of over 12000 news articles from iLur.am, categorized into 7 classes: sport, politics, weather, economy, accidents, art, society. The articles are split into train (2242k tokens) and test sets (425k tokens). Source: iLur News Texts

ImageNet C-OOD (class-out-of-distribution)

This dataset was presented as part of the ICLR 2023 paper 𝘈 𝘧𝘳𝘢𝘮𝘦𝘸𝘰𝘳𝘬 𝘧𝘰𝘳 𝘣𝘦𝘯𝘤𝘩𝘮𝘢𝘳𝘬𝘪𝘯𝘨 𝘊𝘭𝘢𝘴𝘴-𝘰𝘶𝘵-𝘰𝘧-𝘥𝘪𝘴𝘵𝘳𝘪𝘣𝘶𝘵𝘪𝘰𝘯 𝘥𝘦𝘵𝘦𝘤𝘵𝘪𝘰𝘯 𝘢𝘯𝘥 𝘪𝘵𝘴 𝘢𝘱𝘱𝘭𝘪𝘤𝘢𝘵𝘪𝘰𝘯 𝘵𝘰 𝘐𝘮𝘢𝘨𝘦𝘕𝘦𝘵. It is a framework that, based on this dataset (a subset of the ImageNet-21k dataset) is able to generate a C-OOD (AKA open-set recognition) benchmark that covers a variety of difficulty levels. these benchmarks are tailored to the evaluated model. This approach provides a more accurate representation of the model’s own performance. The resulting difficulty levels of our framework allow benchmarking with respect to the difficulty levels most relevant to the task. For example, for a task with a high tolerance for risk (e.g., a task for an entertainment application), the performance of a model on a median difficulty level might be more important than on the hardest difficulty level (severity 10). The opposite might be true for some applications with a low tolerance for risk (e.g., medical applications), for which one requires the best performance to be attained even if the OOD is very hard to detect (severity 10). The paper in which the framework was introduced showed that detection algorithms do not always improve performance on all inputs equally, and could even hurt performance for specific difficulty levels and models. Choosing the combination of (model, detection algorithm) based only on the detection performance on all data may yield sub-optimal results for our specific desired level of difficulty.

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