19,997 machine learning datasets
19,997 dataset results
Click to add a brief description of the dataset (Markdown and LaTeX enabled).
We present a major update upon the original version of Deep Fashion 3D dataset . Compared with the previous version of Deep Fashion3D dataset, Deep Fashion3D V2 is futher equipped with: (1) detailed registered garment meshes with category-specific triangulation, (2) high-resolution texture maps (2048 X 2048 px), (3) more precise and accurate feature line annotations, and (4) garment SMPL pose parameters. We believe this update will further boost the research in high-quality, real-world garment modeling and generation.
A collection of natural language prompt-completion pairs pertaining to multiple-choice Q&A on benchmark tasks based on US census products. Benchmark tasks are made available through a python package dubbed folktexts. The main goal is to serve as a basis to evaluate LLMs' capabilities of uncertainty quantification on uncertain outcomes, i.e., evaluating quantification of aleatoric uncertainty. This is essentially a natural-language version of the popular folktables tabular data package.
Click to add a brief description of the dataset (Markdown and LaTeX enabled).
Revealing latent structure in data is an active field of research, having introduced exciting technologies such as variational autoencoders and adversarial networks, and is essential to push machine learning towards unsupervised knowledge discovery. However, a major challenge is the lack of suitable benchmarks for an objective and quantitative evaluation of learned representations. To address this issue we introduce Morpho-MNIST, a framework that aims to answer: "to what extent has my model learned to represent specific factors of variation in the data?" We extend the popular MNIST dataset by adding a morphometric analysis enabling quantitative comparison of trained models, identification of the roles of latent variables, and characterisation of sample diversity. We further propose a set of quantifiable perturbations to assess the performance of unsupervised and supervised methods on challenging tasks such as outlier detection and domain adaptation. Data and code are available at https
Language models (LMs) as conversational assistants recently became popular tools that help people accomplish a variety of tasks. These typically result from adapting LMs pretrained on general domain text sequences through further instruction-tuning and possibly preference optimisation methods. The evaluation of such LMs would ideally be performed using human judgement, however, this is not scalable. On the other hand, automatic evaluation featuring auxiliary LMs as judges and/or knowledge-based tasks is scalable but struggles with assessing conversational ability and adherence to instructions. To help accelerate the development of LMs as conversational assistants, we propose a novel automatic evaluation task: HumanRankEval (HRE). It consists of a large-scale, diverse and high-quality set of questions, each with several answers authored and scored by humans. To perform evaluation, HRE ranks these answers based on their log-likelihood under the LM’s distribution, and subsequently calcula
The YCB-Ev dataset contains synchronized RGB-D frames and event data that enables evaluating 6DoF object pose estimation algorithms using these modalities. This dataset provides ground truth 6DoF object poses for the same 21 YCB objects that were used in the YCB-Video (YCB-V) dataset, allowing for cross-dataset algorithm performance evaluation. The dataset consists of 21 synchronized event and RGB-D sequences, totalling 13,851 frames (7 minutes and 43 seconds of event data). Notably, 12 of these sequences feature the same object arrangement as the YCB-V subset used in the BOP challenge.
The researchers collected a dataset of 3,500 images of Tilapia fish in a small bowl containing three fish per image. These images were manually annotated using Roboflow, with four keypoints labeled on each fish: mouth, peduncle, belly, and back. While only the mouth and peduncle keypoints were needed for length measurement, the additional keypoints were included to support potential future research using girth for weight determination. The dataset was used to train YOLOv8 models for both keypoint detection and fish counting tasks. For real-world validation, an additional test set of 100 frame pairs (200 images total) captured from two cameras at different angles in actual fish farm conditions was also used.
The researchers collected 3,500 images of Tilapia fish, with each image containing three fish in a small bowl. These images were manually annotated using Roboflow, a tool for creating and managing annotated datasets. Four keypoints were labeled on each fish: mouth, peduncle, belly, and back. While the primary goal was to measure fish length using the mouth and peduncle points, the additional keypoints (belly and back) were included to support potential future research, such as using girth to determine fish weight. This dataset was used to train a YOLOv8 model for keypoint detection, achieving high accuracy in identifying these crucial points on the Tilapia fish.
This dataset is used to examine the influence of image memorability on social media virality.
A novel audio-visual mouse saliency (AViMoS) dataset with the following key-features:
This dataset is a supplement to the github repositiry (https://github.com/pfilonenko/ML_for_TwoSampleTesting) and paper addressed to solve the two-sample problem under right-censored observations using Machine Learning. The problem statement can be formualted as H0: S1(t)=S2(t) versus H: S1(t)≠S_2(t) where S1(t) and S2(t) are survival functions of samples X1 and X2.
Source: Radar reflectivity data from the HydroMeteorological Service of Arpae (Emilia-Romagna, Italy). Geographical Coverage: Emilia-Romagna region, including flat Po Valley, the Apennines, and coastal areas. Time Period: 6 years (2015–2020). Data Resolution: Temporal: Scans taken every 5 minutes. Spatial: 1 km grid resolution. Area covered: 125 km radius per scan, covering a total of 71,172 square km. Reflectivity Range: 0 to 60 dBZ, clipped from an original range of -20 dBZ to 60 dBZ. Total Time Steps: 630,720 time steps in total. Precipitating Events: 179,264 time steps representing precipitating sequences. Non-precipitating Data: 71.5% of the data was discarded (non-precipitating). Dataset Split: -Training: 149,524 time steps. -Validation: 7,869 time steps. Test Sets: -Tokenizer Test Set (TTS): 21,871 radar images, focusing on extreme events. -Forecaster Test Set (FTS): 1,450 time steps from 10 selected extreme weather events (12 hours each). Data Augmentation: Random cropping, 90-
Click to add a brief description of the dataset (Markdown and LaTeX enabled).
FineCops-Ref is a dataset for Compositional Referring Expression Comprehension (REC) that rigorously evaluates Vision-Language Models (VLMs) on compositional reasoning and their ability to identify inconsistencies between images and text. Beyond standard REC tasks, it challenges models with fine-grained correspondences involving objects, attributes, and relationships. The dataset comprises both training and testing sets, designed to thoroughly assess model performance across various difficulty level.
Dataset to benchmark Continual Learning for Object Detection in a Tiny Robotics settings.
The Amsterdam Open MRI Collection (AOMIC) is a collection of three datasets with multimodal (3T) MRI data including structural (T1-weighted), diffusion-weighted, and (resting-state and task-based) functional BOLD MRI data, as well as detailed demographics and psychometric variables from a large set of healthy participants (N = 928, N = 226, and N = 216). Notably, task-based fMRI was collected during various robust paradigms (targeting naturalistic vision, emotion perception, working memory, face perception, cognitive conflict and control, and response inhibition) for which extensively annotated event-files are available. For each dataset and data modality, we provide the data in both raw and preprocessed form (both compliant with the Brain Imaging Data Structure), which were subjected to extensive (automated and manual) quality control.
a large-scale, slow event-related human fMRI study incorporating 5,000 real-world images as stimuli. BOLD5000 is an order of magnitude larger than any extant slow event-related fMRI dataset, with ∼20 hours of MRI scanning per each of four participants. By scaling the size of the image dataset used in fMRI, we hope to facilitate greater integration between the fields of human and computer vision. To that end, BOLD5000 also uniquely uses images drawn from the three most commonly-used computer vision datasets: SUN, COCO, and ImageNet. Beyond standard fMRI analysis techniques, we use both representational similarity analysis14 and, uniquely, t-distributed stochastic neighbor embedding visualizations15, to validate the quality of our data. In sum, we hope that BOLD5000 engenders greater collaboration between the two fields of vision science, fulfilling Marr’s dream.
The Courtois NeuroMod project aims at training artificial neural networks using extensive experimental data on individual human brain activity and behaviour. Courtois NeuroMod data are freely shared with the scientific community to advance research at the interface of neuroscience and artificial intelligence.
This is an extension to the studyforrest dataset – a versatile resource for studying the behavior of the human brain in situations of real-life complexity (http://studyforrest.org). This release adds more high-resolution, ultra high-field (7 Tesla) functional magnetic resonance imaging (fMRI) data from the same individuals. The twenty participants were repeatedly stimulated with a total of 25 music clips, with and without speech content, from five different genres using a slow event-related paradigm. The data release includes raw fMRI data, as well as precomputed structural alignments for within-subject and group analysis. In addition to fMRI, simultaneously recorded cardiac and respiratory traces, as well the complete implementation of the stimulation paradigm, including stimuli, are provided. An initial quality control analysis reveals distinguishable patterns of response to individual genres throughout a large expanse of areas known to be involved in auditory and speech processing.