The Large-Scale AI Models database documents over 200 models trained with more than 10²³ floating point operations, at the leading edge of scale and capabilities.

The training dataset consisting of 20 million pairs of product offers referring to the same products. The offers were extracted from 43 thousand e-shops which provide schema.org annotations including some form of product ID such as a GTIN or MPN. We also created a gold standard by manually verifying 2000 pairs of offers belonging to four different product categories.

The MNIST Large Scale dataset is based on the classic MNIST dataset, but contains large scale variations up to a factor of 16. The motivation behind creating this dataset was to enable testing the ability of different algorithms to learn in the presence of large scale variability and specifically the ability to generalise to new scales not present in the training set over wide scale ranges.

The dataset contains training data for each one of the relative size factors 1, 2 and 4 relative to the original MNIST dataset and testing data for relative scaling factors between 1/2 and 8, with a ratio of $\sqrt[4]{2}$ between adjacent scales.

The U.S. AI Training Dataset Market size was valued at USD 590.4 million in 2023 and is projected to reach USD 1880.70 million by 2032, exhibiting a CAGR of 18.0 % during the forecasts period. The U. S. AI training dataset market deals with the generation, selection, and organization of datasets used in training artificial intelligence. These datasets contain the requisite information that the machine learning algorithms need to infer and learn from. Conducts include the advancement and improvement of AI solutions in different fields of business like transport, medical analysis, computing language, and money related measurements. The applications include training the models for activities such as image classification, predictive modeling, and natural language interface. Other emerging trends are the change in direction of more and better-quality, various and annotated data for the improvement of model efficiency, synthetic data generation for data shortage, and data confidentiality and ethical issues in dataset management. Furthermore, due to arising technologies in artificial intelligence and machine learning, there is a noticeable development in building and using the datasets. Recent developments include: In February 2024, Google struck a deal worth USD 60 million per year with Reddit that will give the former real-time access to the latter’s data and use Google AI to enhance Reddit’s search capabilities. , In February 2024, Microsoft announced around USD 2.1 billion investment in Mistral AI to expedite the growth and deployment of large language models. The U.S. giant is expected to underpin Mistral AI with Azure AI supercomputing infrastructure to provide top-notch scale and performance for AI training and inference workloads. .

One of the key problems that arises in many areas is to estimate a potentially nonlinear function [tex] G(x, \theta)[/tex] given input and output samples tex [/tex] so that [tex]y approx G(x, \theta)[/tex]. There are many approaches to addressing this regression problem. Neural networks, regression trees, and many other methods have been developed to estimate [tex]$G$[/tex] given the input output pair tex [/tex]. One method that I have worked with is called Gaussian process regression. There many good texts and papers on the subject. For more technical information on the method and its applications see: http://www.gaussianprocess.org/ A key problem that arises in developing these models on very large data sets is that it ends up requiring an [tex]O(N^3)[/tex] computation where N is the number of data points and the training sample. Obviously this becomes very problematic when N is large. I discussed this problem with Leslie Foster, a mathematics professor at San Jose State University. He, along with some of his students, developed a method to address this problem based on Cholesky decomposition and pivoting. He also shows that this leads to a numerically stable result. If ou're interested in some light reading, I’d suggest you take a look at his recent paper (which was accepted in the Journal of Machine Learning Research) posted on dashlink. We've also posted code for you to try it out. Let us know how it goes. If you are interested in applications of this method in the area of prognostics, check out our new paper on the subject which was published in IEEE Transactions on Systems, Man, and Cybernetics.

Motivation

The goal of introducing the Rescaled Fashion-MNIST with translations dataset is to provide a dataset that contains scale variations (up to a factor of 4), to evaluate the ability of networks to generalise to scales not present in the training data, and to additionally provide a way to test network object detection and object localisation abilities on image data where the objects are not centred.

The Rescaled Fashion-MNIST with translations dataset was introduced in the paper:

[1] A. Perzanowski and T. Lindeberg (2025) "Scale generalisation properties of extended scale-covariant and scale-invariant Gaussian derivative networks on image datasets with spatial scaling variations”, Journal of Mathematical Imaging and Vision, 67(29), https://doi.org/10.1007/s10851-025-01245-x.

with a pre-print available at arXiv:

[2] Perzanowski and Lindeberg (2024) "Scale generalisation properties of extended scale-covariant and scale-invariant Gaussian derivative networks on image datasets with spatial scaling variations”, arXiv preprint arXiv:2409.11140.

Importantly, the Rescaled Fashion-MNIST with translations dataset is more challenging than the MNIST Large Scale dataset, introduced in:

[3] Y. Jansson and T. Lindeberg (2022) "Scale-invariant scale-channel networks: Deep networks that generalise to previously unseen scales", Journal of Mathematical Imaging and Vision, 64(5): 506-536, https://doi.org/10.1007/s10851-022-01082-2.

Access and rights

The Rescaled Fashion-MNIST with translations dataset is provided on the condition that you provide proper citation for the original Fashion-MNIST dataset:

[4] Xiao, H., Rasul, K., and Vollgraf, R. (2017) “Fashion-MNIST: A novel image dataset for benchmarking machine learning algorithms”, arXiv preprint arXiv:1708.07747

and also for this new rescaled version, using the reference [1] above.

The data set is made available on request. If you would be interested in trying out this data set, please make a request in the system below, and we will grant you access as soon as possible.

The dataset

The Rescaled FashionMNIST with translations dataset is generated by rescaling 28×28 gray-scale images of clothes from the original FashionMNIST dataset [4]. The scale variations are up to a factor of 4, and the images are embedded within black images of size 72x72. The objects within the images have also been randomly shifted in the spatial domain, with the object always at least 4 pixels away from the image boundary. The imresize() function in Matlab was used for the rescaling, with default anti-aliasing turned on, and bicubic interpolation overshoot removed by clipping to the [0, 255] range. The details of how the dataset was created can be found in [1].

There are 10 different classes in the dataset: “T-shirt/top”, “trouser”, “pullover”, “dress”, “coat”, “sandal”, “shirt”, “sneaker”, “bag” and “ankle boot”. In the dataset, these are represented by integer labels in the range [0, 9].

The dataset is split into 50 000 training samples, 10 000 validation samples and 10 000 testing samples. The training dataset is generated using the initial 50 000 samples from the original Fashion-MNIST training set. The validation dataset, on the other hand, is formed from the final 10 000 images of that same training set. For testing, all test datasets are built from the 10 000 images contained in the original Fashion-MNIST test set.

The h5 files containing the dataset

The training dataset file (~2.9 GB) for scale 1, which also contains the corresponding validation and test data for the same scale, is:

fashionmnist_with_scale_variations_and_translations_tr50000_vl10000_te10000_outsize72-72_scte1p000_scte1p000.h5

Additionally, for the Rescaled FashionMNIST with translations dataset, there are 9 datasets (~415 MB each) for testing scale generalisation at scales not present in the training set. Each of these datasets is rescaled using a different image scaling factor, 2^k/4, with k being integers in the range [-4, 4]:

fashionmnist_with_scale_variations_and_translations_te10000_outsize72-72_scte0p500.h5
fashionmnist_with_scale_variations_and_translations_te10000_outsize72-72_scte0p595.h5
fashionmnist_with_scale_variations_and_translations_te10000_outsize72-72_scte0p707.h5
fashionmnist_with_scale_variations_and_translations_te10000_outsize72-72_scte0p841.h5
fashionmnist_with_scale_variations_and_translations_te10000_outsize72-72_scte1p000.h5
fashionmnist_with_scale_variations_and_translations_te10000_outsize72-72_scte1p189.h5
fashionmnist_with_scale_variations_and_translations_te10000_outsize72-72_scte1p414.h5
fashionmnist_with_scale_variations_and_translations_te10000_outsize72-72_scte1p682.h5
fashionmnist_with_scale_variations_and_translations_te10000_outsize72-72_scte2p000.h5

These dataset files were used for the experiments presented in Figure 8 in [1].

Instructions for loading the data set

The datasets are saved in HDF5 format, with the partitions in the respective h5 files named as
('/x_train', '/x_val', '/x_test', '/y_train', '/y_test', '/y_val'); which ones exist depends on which data split is used.

The training dataset can be loaded in Python as:

with h5py.File(`

x_train = np.array( f["/x_train"], dtype=np.float32)
x_val = np.array( f["/x_val"], dtype=np.float32)
x_test = np.array( f["/x_test"], dtype=np.float32)
y_train = np.array( f["/y_train"], dtype=np.int32)
y_val = np.array( f["/y_val"], dtype=np.int32)
y_test = np.array( f["/y_test"], dtype=np.int32)

We also need to permute the data, since Pytorch uses the format [num_samples, channels, width, height], while the data is saved as [num_samples, width, height, channels]:

x_train = np.transpose(x_train, (0, 3, 1, 2))
x_val = np.transpose(x_val, (0, 3, 1, 2))
x_test = np.transpose(x_test, (0, 3, 1, 2))

The test datasets can be loaded in Python as:

with h5py.File(`

x_test = np.array( f["/x_test"], dtype=np.float32)
y_test = np.array( f["/y_test"], dtype=np.int32)

The test datasets can be loaded in Matlab as:

x_test = h5read(`

The images are stored as [num_samples, x_dim, y_dim, channels] in HDF5 files. The pixel intensity values are not normalised, and are in a [0, 255] range.

There is also a closely related Fashion-MNIST dataset, which in addition to scaling variations keeps the objects in the frame centred, meaning no spatial translations are used.

Artificial Intelligence (AI) Training Dataset Market Outlook

According to our latest research, the global Artificial Intelligence (AI) Training Dataset market size reached USD 3.15 billion in 2024, reflecting robust industry momentum. The market is expanding at a notable CAGR of 20.8% and is forecasted to attain USD 20.92 billion by 2033. This impressive growth is primarily attributed to the surging demand for high-quality, annotated datasets to fuel machine learning and deep learning models across diverse industry verticals. The proliferation of AI-driven applications, coupled with rapid advancements in data labeling technologies, is further accelerating the adoption and expansion of the AI training dataset market globally.

One of the most significant growth factors propelling the AI training dataset market is the exponential rise in data-driven AI applications across industries such as healthcare, automotive, retail, and finance. As organizations increasingly rely on AI-powered solutions for automation, predictive analytics, and personalized customer experiences, the need for large, diverse, and accurately labeled datasets has become critical. Enhanced data annotation techniques, including manual, semi-automated, and fully automated methods, are enabling organizations to generate high-quality datasets at scale, which is essential for training sophisticated AI models. The integration of AI in edge devices, smart sensors, and IoT platforms is further amplifying the demand for specialized datasets tailored for unique use cases, thereby fueling market growth.

Another key driver is the ongoing innovation in machine learning and deep learning algorithms, which require vast and varied training data to achieve optimal performance. The increasing complexity of AI models, especially in areas such as computer vision, natural language processing, and autonomous systems, necessitates the availability of comprehensive datasets that accurately represent real-world scenarios. Companies are investing heavily in data collection, annotation, and curation services to ensure their AI solutions can generalize effectively and deliver reliable outcomes. Additionally, the rise of synthetic data generation and data augmentation techniques is helping address challenges related to data scarcity, privacy, and bias, further supporting the expansion of the AI training dataset market.

The market is also benefiting from the growing emphasis on ethical AI and regulatory compliance, particularly in data-sensitive sectors like healthcare, finance, and government. Organizations are prioritizing the use of high-quality, unbiased, and diverse datasets to mitigate algorithmic bias and ensure transparency in AI decision-making processes. This focus on responsible AI development is driving demand for curated datasets that adhere to strict quality and privacy standards. Moreover, the emergence of data marketplaces and collaborative data-sharing initiatives is making it easier for organizations to access and exchange valuable training data, fostering innovation and accelerating AI adoption across multiple domains.

From a regional perspective, North America currently dominates the AI training dataset market, accounting for the largest revenue share in 2024, driven by significant investments in AI research, a mature technology ecosystem, and the presence of leading AI companies and data annotation service providers. Europe and Asia Pacific are also witnessing rapid growth, with increasing government support for AI initiatives, expanding digital infrastructure, and a rising number of AI startups. While North America sets the pace in terms of technological innovation, Asia Pacific is expected to exhibit the highest CAGR during the forecast period, fueled by the digital transformation of emerging economies and the proliferation of AI applications across various industry sectors.

Data Type Analysis

The AI training dataset market is segmented by data type into Text, Image/Video, Audio, and Others, each playing a crucial role in powering different AI applications. Text da

Datasets required for supervised fine-tuning. Only sol_20n_1280000t_?.npy chunks are required for model scaling (20n for 20-nodes), the remainder are needed for TSP node scaling.Recent work on neural scaling laws demonstrates that model performance scales predictably with compute budget, model size, and dataset size. In this work, we develop scaling laws based on problem complexity. We analyze two fundamental complexity measures: solution space size and representation space size. Using the Traveling Salesman Problem (TSP) as a case study, we show that combinatorial optimization promotes smooth cost trends, and therefore meaningful scaling laws can be obtained even in the absence of an interpretable loss. We then show that suboptimality grows predictably for fixed-size models when scaling the number of TSP nodes or spatial dimensions, independent of whether the model was trained with reinforcement learning or supervised fine-tuning on a static dataset. We conclude with an analogy to problem complexity scaling in local search, showing that a much simpler gradient descent of the cost landscape produces similar trends.

The implemented gym-hpa is a custom OpenAi Gym environment for the training of Reinforcement Learning (RL) agents for auto-scaling research in the Kubernetes (K8s) platform.

Two environments exist based on the Redis Cluster and Online Boutique applications.

Two collected datasets are shared here. The code has been released here: https://github.com/jpedro1992/gym-hpa

Related Publication: Santos, J. et al. "gym-hpa: Efficient auto-scaling via reinforcement learning for complex microservice-based applications in Kubernetes." NOMS2023, the IEEE/IFIP Network Operations and Management Symposium. 2023.

AI Training Dataset Market Outlook

The global AI training dataset market size was valued at approximately USD 1.2 billion in 2023 and is projected to reach USD 6.5 billion by 2032, growing at a compound annual growth rate (CAGR) of 20.5% from 2024 to 2032. This substantial growth is driven by the increasing adoption of artificial intelligence across various industries, the necessity for large-scale and high-quality datasets to train AI models, and the ongoing advancements in AI and machine learning technologies.

One of the primary growth factors in the AI training dataset market is the exponential increase in data generation across multiple sectors. With the proliferation of internet usage, the expansion of IoT devices, and the digitalization of industries, there is an unprecedented volume of data being generated daily. This data is invaluable for training AI models, enabling them to learn and make more accurate predictions and decisions. Moreover, the need for diverse and comprehensive datasets to improve AI accuracy and reliability is further propelling market growth.

Another significant factor driving the market is the rising investment in AI and machine learning by both public and private sectors. Governments around the world are recognizing the potential of AI to transform economies and improve public services, leading to increased funding for AI research and development. Simultaneously, private enterprises are investing heavily in AI technologies to gain a competitive edge, enhance operational efficiency, and innovate new products and services. These investments necessitate high-quality training datasets, thereby boosting the market.

The proliferation of AI applications in various industries, such as healthcare, automotive, retail, and finance, is also a major contributor to the growth of the AI training dataset market. In healthcare, AI is being used for predictive analytics, personalized medicine, and diagnostic automation, all of which require extensive datasets for training. The automotive industry leverages AI for autonomous driving and vehicle safety systems, while the retail sector uses AI for personalized shopping experiences and inventory management. In finance, AI assists in fraud detection and risk management. The diverse applications across these sectors underline the critical need for robust AI training datasets.

As the demand for AI applications continues to grow, the role of Ai Data Resource Service becomes increasingly vital. These services provide the necessary infrastructure and tools to manage, curate, and distribute datasets efficiently. By leveraging Ai Data Resource Service, organizations can ensure that their AI models are trained on high-quality and relevant data, which is crucial for achieving accurate and reliable outcomes. The service acts as a bridge between raw data and AI applications, streamlining the process of data acquisition, annotation, and validation. This not only enhances the performance of AI systems but also accelerates the development cycle, enabling faster deployment of AI-driven solutions across various sectors.

Regionally, North America currently dominates the AI training dataset market due to the presence of major technology companies and extensive R&D activities in the region. However, Asia Pacific is expected to witness the highest growth rate during the forecast period, driven by rapid technological advancements, increasing investments in AI, and the growing adoption of AI technologies across various industries in countries like China, India, and Japan. Europe and Latin America are also anticipated to experience significant growth, supported by favorable government policies and the increasing use of AI in various sectors.

Data Type Analysis

The data type segment of the AI training dataset market encompasses text, image, audio, video, and others. Each data type plays a crucial role in training different types of AI models, and the demand for specific data types varies based on the application. Text data is extensively used in natural language processing (NLP) applications such as chatbots, sentiment analysis, and language translation. As the use of NLP is becoming more widespread, the demand for high-quality text datasets is continually rising. Companies are investing in curated text datasets that encompass diverse languages and dialects to improve the accuracy and efficiency of NLP models.

Image data is critical for computer vision application

Current moist physics parameterization schemes in general circulation models (GCMs) are the main source of biases in simulated precipitation and atmospheric circulation. Recent advances in machine learning make it possible to explore data-driven approaches to developing parameterization for moist physics processes such as convection and clouds. This study aims to develop a new moist physics parameterization scheme based on deep learning. We use a residual convolutional neural network (ResNet) for this purpose. It is trained with one-year simulation from a superparameterized GCM, SPCAM. An independent year of SPCAM simulation is used for evaluation. In the design of the neural network, referred to as ResCu, the moist static energy conservation during moist processes is considered. In addition, the past history of the atmospheric states, convection and clouds are also considered. The predicted variables from the neural network are GCM grid-scale heating and drying rates by convection and clouds, and cloud liquid and ice water contents. Precipitation is derived from predicted moisture tendency. In the independent-data test, ResCu can accurately reproduce the SPCAM simulation in both time-mean and temporal variance. Comparison with other neural networks demonstrates the superior performance of ResNet architecture. ResCu is further tested in a single column model for both continental midlatitude warm season convection and tropical monsoonal convection. In both cases, it simulates the timing and intensity of convective events well. In the prognostic test of tropical convection case, the simulated temperature and moisture biases with ResCu are smaller than those using conventional convection and cloud parameterizations. This dataset is extracted from a simulation using a Superparameterized GCM, SPCAM (https://wiki.ucar.edu/display/ccsm/Superparameterized+CAM+(SPCAM)). The SPCAM implements a 2-D CRM in CAM5.2 to replace its conventional parameterization for moist convection and large-scale condensation. The dynamic framework of CAM5 has a horizontal resolution of 1.9x2.5 degrees and 30 vertical levels that are shared with the embedded CRM. The SPCAM used in this study has a coupled land surface model Community Land Model 4.0 (CLM4.0). It uses a prescribed climatological sea surface temperature field that comes with the CAM5 model. It is run for three years and 4 months from Jan. 1st in 1998 to March 31st in 2001 with a time step of 20 minutes. The first year and 4 months are for spin up, the second year is used for training and the third year is used for testing and evaluation. The training data from SPCAM is output every timestep. This dataset contains one year training data and one year evaluation data. The training samples of the entire year (from yr-2 of simulation) are compressed in SPCAM_ML_Han_et_al_0.tar.gz, and testing samples of the entire year (from yr-3 of simulation) are compressed in SPCAM_ML_Han_et_al_1.tar.gz. In each dataset, there are a data documentation file and 365 netCDF data files (one file for each day) that are marked by its date. The variable fields contain temperature and moisture tendencies and cloud water and cloud ice from the CRM, and vertical profiles of temperature and moisture and large-scale temperature and moisture tendencies from the dynamic core of SPCAM’s host model CAM5 and PBL diffusion. In addition, we include surface sensible and latent heat fluxes. For more details, please read the data documentation inside the tar.gz files.

The global Artificial Intelligence (AI) Training Dataset market is projected to reach $1605.2 million by 2033, exhibiting a CAGR of 9.4% from 2025 to 2033. The surge in demand for AI training datasets is driven by the increasing adoption of AI and machine learning technologies in various industries such as healthcare, financial services, and manufacturing. Moreover, the growing need for reliable and high-quality data for training AI models is further fueling the market growth. Key market trends include the increasing adoption of cloud-based AI training datasets, the emergence of synthetic data generation, and the growing focus on data privacy and security. The market is segmented by type (image classification dataset, voice recognition dataset, natural language processing dataset, object detection dataset, and others) and application (smart campus, smart medical, autopilot, smart home, and others). North America is the largest regional market, followed by Europe and Asia Pacific. Key companies operating in the market include Appen, Speechocean, TELUS International, Summa Linguae Technologies, and Scale AI. Artificial Intelligence (AI) training datasets are critical for developing and deploying AI models. These datasets provide the data that AI models need to learn, and the quality of the data directly impacts the performance of the model. The AI training dataset market landscape is complex, with many different providers offering datasets for a variety of applications. The market is also rapidly evolving, as new technologies and techniques are developed for collecting, labeling, and managing AI training data.

The accuracy of each annotation method with respect to the expert annotations in each dataset. Aggregating maintains best or near-best accuracy across tasks.

Object detection is a vital part of any autonomous vision system and to obtain a high performing object detector data is needed. The object detection task aims to detect and classify different objects using camera input and getting bounding boxes containing the objects as output. This is usually done by utilizing deep neural networks.

When training an object detector a large amount of data is used, however it is not always practical to collect large amounts of data. This has led to multiple different techniques which decreases the amount of data needed. Examples of such techniques are transfer learning and domain adaptation. Working with construction equipment is a time consuming process and we wanted to examine if it was possible to use scale-model data to train a network and then used that network to detect real objects with no additional training.

This small dataset contains training and validation data of a scale dump truck in different environments while the test set contains images of a full size dump truck of similar model. The aim of the dataset is to train a network to classify wheels, cabs and tipping bodies of a scale-model dump truck and use that to classify the same classes on a full-scale dump truck.

The label structure of the dataset is the YOLO v3 structure, where the classes corresponds to a integer value, such that: Wheel: 0 Cab: 1 Tipping body: 2

The Large-Scale Model Training Machine market is experiencing rapid growth, driven by the increasing demand for sophisticated AI applications across various sectors. The market size in 2025 is estimated at $15 billion, projecting a Compound Annual Growth Rate (CAGR) of 25% from 2025 to 2033. This robust growth is fueled by several key factors, including the proliferation of big data, advancements in deep learning algorithms, and the rising adoption of cloud computing for AI model training. The expansion of edge computing infrastructure also contributes significantly, enabling faster and more efficient training of large-scale models closer to the data source. Major players like Google, Amazon, Microsoft, and others are heavily investing in research and development, further accelerating market expansion. The market segmentation is largely driven by deployment models (on-premises vs. cloud), application domains (image recognition, natural language processing, etc.), and geographical regions. Competition is fierce, with established tech giants and emerging AI startups vying for market share through innovative solutions and strategic partnerships. The continued growth of the Large-Scale Model Training Machine market is expected to be shaped by several emerging trends. These include the increasing adoption of specialized hardware like GPUs and TPUs, the development of more efficient training algorithms, and the growing interest in federated learning for enhanced data privacy. However, challenges remain, such as the high cost of infrastructure and specialized expertise, along with concerns about data security and ethical implications of advanced AI models. Despite these challenges, the long-term outlook for the Large-Scale Model Training Machine market remains extremely positive, with sustained growth predicted well into the next decade, driven by an ever-increasing need for powerful and sophisticated AI capabilities.

Motivation

The goal of introducing the Rescaled Fashion-MNIST dataset is to provide a dataset that contains scale variations (up to a factor of 4), to evaluate the ability of networks to generalise to scales not present in the training data.

The Rescaled Fashion-MNIST dataset was introduced in the paper:

[1] A. Perzanowski and T. Lindeberg (2025) "Scale generalisation properties of extended scale-covariant and scale-invariant Gaussian derivative networks on image datasets with spatial scaling variations”, Journal of Mathematical Imaging and Vision, 67(29), https://doi.org/10.1007/s10851-025-01245-x.

with a pre-print available at arXiv:

[2] Perzanowski and Lindeberg (2024) "Scale generalisation properties of extended scale-covariant and scale-invariant Gaussian derivative networks on image datasets with spatial scaling variations”, arXiv preprint arXiv:2409.11140.

Importantly, the Rescaled Fashion-MNIST dataset is more challenging than the MNIST Large Scale dataset, introduced in:

[3] Y. Jansson and T. Lindeberg (2022) "Scale-invariant scale-channel networks: Deep networks that generalise to previously unseen scales", Journal of Mathematical Imaging and Vision, 64(5): 506-536, https://doi.org/10.1007/s10851-022-01082-2.

Access and rights

The Rescaled Fashion-MNIST dataset is provided on the condition that you provide proper citation for the original Fashion-MNIST dataset:

[4] Xiao, H., Rasul, K., and Vollgraf, R. (2017) “Fashion-MNIST: A novel image dataset for benchmarking machine learning algorithms”, arXiv preprint arXiv:1708.07747

and also for this new rescaled version, using the reference [1] above.

The data set is made available on request. If you would be interested in trying out this data set, please make a request in the system below, and we will grant you access as soon as possible.

The dataset

The Rescaled FashionMNIST dataset is generated by rescaling 28×28 gray-scale images of clothes from the original FashionMNIST dataset [4]. The scale variations are up to a factor of 4, and the images are embedded within black images of size 72x72, with the object in the frame always centred. The imresize() function in Matlab was used for the rescaling, with default anti-aliasing turned on, and bicubic interpolation overshoot removed by clipping to the [0, 255] range. The details of how the dataset was created can be found in [1].

There are 10 different classes in the dataset: “T-shirt/top”, “trouser”, “pullover”, “dress”, “coat”, “sandal”, “shirt”, “sneaker”, “bag” and “ankle boot”. In the dataset, these are represented by integer labels in the range [0, 9].

The dataset is split into 50 000 training samples, 10 000 validation samples and 10 000 testing samples. The training dataset is generated using the initial 50 000 samples from the original Fashion-MNIST training set. The validation dataset, on the other hand, is formed from the final 10 000 images of that same training set. For testing, all test datasets are built from the 10 000 images contained in the original Fashion-MNIST test set.

The h5 files containing the dataset

The training dataset file (~2.9 GB) for scale 1, which also contains the corresponding validation and test data for the same scale, is:

fashionmnist_with_scale_variations_tr50000_vl10000_te10000_outsize72-72_scte1p000_scte1p000.h5

Additionally, for the Rescaled FashionMNIST dataset, there are 9 datasets (~415 MB each) for testing scale generalisation at scales not present in the training set. Each of these datasets is rescaled using a different image scaling factor, 2^k/4, with k being integers in the range [-4, 4]:

fashionmnist_with_scale_variations_te10000_outsize72-72_scte0p500.h5
fashionmnist_with_scale_variations_te10000_outsize72-72_scte0p595.h5
fashionmnist_with_scale_variations_te10000_outsize72-72_scte0p707.h5
fashionmnist_with_scale_variations_te10000_outsize72-72_scte0p841.h5
fashionmnist_with_scale_variations_te10000_outsize72-72_scte1p000.h5
fashionmnist_with_scale_variations_te10000_outsize72-72_scte1p189.h5
fashionmnist_with_scale_variations_te10000_outsize72-72_scte1p414.h5
fashionmnist_with_scale_variations_te10000_outsize72-72_scte1p682.h5
fashionmnist_with_scale_variations_te10000_outsize72-72_scte2p000.h5

These dataset files were used for the experiments presented in Figures 6, 7, 14, 16, 19 and 23 in [1].

Instructions for loading the data set

The datasets are saved in HDF5 format, with the partitions in the respective h5 files named as
('/x_train', '/x_val', '/x_test', '/y_train', '/y_test', '/y_val'); which ones exist depends on which data split is used.

The training dataset can be loaded in Python as:

with h5py.File(`

x_train = np.array( f["/x_train"], dtype=np.float32)
x_val = np.array( f["/x_val"], dtype=np.float32)
x_test = np.array( f["/x_test"], dtype=np.float32)
y_train = np.array( f["/y_train"], dtype=np.int32)
y_val = np.array( f["/y_val"], dtype=np.int32)
y_test = np.array( f["/y_test"], dtype=np.int32)

We also need to permute the data, since Pytorch uses the format [num_samples, channels, width, height], while the data is saved as [num_samples, width, height, channels]:

x_train = np.transpose(x_train, (0, 3, 1, 2))
x_val = np.transpose(x_val, (0, 3, 1, 2))
x_test = np.transpose(x_test, (0, 3, 1, 2))

The test datasets can be loaded in Python as:

with h5py.File(`

x_test = np.array( f["/x_test"], dtype=np.float32)
y_test = np.array( f["/y_test"], dtype=np.int32)

The test datasets can be loaded in Matlab as:

x_test = h5read(`

The images are stored as [num_samples, x_dim, y_dim, channels] in HDF5 files. The pixel intensity values are not normalised, and are in a [0, 255] range.

There is also a closely related Fashion-MNIST with translations dataset, which in addition to scaling variations also comprises spatial translations of the objects.

Many e-shops have started to mark-up product data within their HTML pages using the schema.org vocabulary. The Web Data Commons project regularly extracts such data from the Common Crawl, a large public web crawl. The Web Data Commons Training and Test Sets for Large-Scale Product Matching contain product offers from different e-shops in the form of binary product pairs (with corresponding label "match" or "no match") for four product categories, computers, cameras, watches and shoes.

In order to support the evaluation of machine learning-based matching methods, the data is split into training, validation and test sets. For each product category, we provide training sets in four different sizes (2.000-70.000 pairs). Furthermore there are sets of ids for each training set for a possible validation split (stratified random draw) available. The test set for each product category consists of 1.100 product pairs. The labels of the test sets were manually checked while those of the training sets were derived using shared product identifiers from the Web via weak supervision.

The data stems from the WDC Product Data Corpus for Large-Scale Product Matching - Version 2.0 which consists of 26 million product offers originating from 79 thousand websites.

This dataset is for a study evaluating Rat Grimace Scale scores generated by novice raters. Four raters scored 2 image banks: S1 (n = 42 images) and S2 (n = 150 images). The S2 image bank was scored twice. Scoring was performed independently, with a facilitated discussion following each training session, during which raters could discuss images that were difficult to score.

The global Artificial Intelligence (AI) Training Dataset market is experiencing robust growth, driven by the increasing adoption of AI across diverse sectors. The market's expansion is fueled by the burgeoning need for high-quality data to train sophisticated AI algorithms capable of powering applications like smart campuses, autonomous vehicles, and personalized healthcare solutions. The demand for diverse dataset types, including image classification, voice recognition, natural language processing, and object detection datasets, is a key factor contributing to market growth. While the exact market size in 2025 is unavailable, considering a conservative estimate of a $10 billion market in 2025 based on the growth trend and reported market sizes of related industries, and a projected CAGR (Compound Annual Growth Rate) of 25%, the market is poised for significant expansion in the coming years. Key players in this space are leveraging technological advancements and strategic partnerships to enhance data quality and expand their service offerings. Furthermore, the increasing availability of cloud-based data annotation and processing tools is further streamlining operations and making AI training datasets more accessible to businesses of all sizes. Growth is expected to be particularly strong in regions with burgeoning technological advancements and substantial digital infrastructure, such as North America and Asia Pacific. However, challenges such as data privacy concerns, the high cost of data annotation, and the scarcity of skilled professionals capable of handling complex datasets remain obstacles to broader market penetration. The ongoing evolution of AI technologies and the expanding applications of AI across multiple sectors will continue to shape the demand for AI training datasets, pushing this market toward higher growth trajectories in the coming years. The diversity of applications—from smart homes and medical diagnoses to advanced robotics and autonomous driving—creates significant opportunities for companies specializing in this market. Maintaining data quality, security, and ethical considerations will be crucial for future market leadership.

This repository contains part 7/7 of the full dataset used for the models of the preprint "Can AI be enabled to dynamical downscaling? Training a Latent Diffusion Model to mimic km-scale COSMO-CLM downscaling of ERA5 over Italy".

This dataset comprises 3 years of normalized hourly data for both low-resolution predictors [16 km] and high-resolution target variables [2km] (2mT and 10-m U and V), from 2018-2019. Low-resolution data are preprocessed ERA5 data while high-resolution data are preprocessed VHR-REA CMCC data. Details on the performed preprocessing are available in the paper.

To use the data, clone the corresponding repository, unzip this zip file in the data folder, and download from Zenodo the other parts of the dataset listed in the related works.