Abstract

In the last years, neural networks have evolved from laboratory environments to the state-of-the-art for many real-world problems. Our hypothesis is that neural network models (i.e., their weights and biases) evolve on unique, smooth trajectories in weight space during training. Following, a population of such neural network models (refereed to as “model zoo”) would form topological structures in weight space. We think that the geometry, curvature and smoothness of these structures contain information about the state of training and can be reveal latent properties of individual models. With such zoos, one could investigate novel approaches for (i) model analysis, (ii) discover unknown learning dynamics, (iii) learn rich representations of such populations, or (iv) exploit the model zoos for generative modelling of neural network weights and biases. Unfortunately, the lack of standardized model zoos and available benchmarks significantly increases the friction for further research about populations of neural networks. With this work, we publish a novel dataset of model zoos containing systematically generated and diverse populations of neural network models for further research. In total the proposed model zoo dataset is based on six image datasets, consist of 24 model zoos with varying hyperparameter combinations are generated and includes 47’360 unique neural network models resulting in over 2’415’360 collected model states. Additionally, to the model zoo data we provide an in-depth analysis of the zoos and provide benchmarks for multiple downstream tasks as mentioned before.

Dataset

This dataset is part of a larger collection of model zoos and contains the zoos trained on the labelled samples from Fashion-MNIST. All zoos with extensive information and code can be found at www.modelzoos.cc.

This repository contains two types of files: the raw model zoos as collections of models (file names beginning with "fmnist_"), as well as preprocessed model zoos wrapped in a custom pytorch dataset class (filenames beginning with "dataset"). Zoos are trained in three configurations varying the seed only (seed), varying hyperparameters with fixed seeds (hyp_fix) or varying hyperparameters with random seeds (hyp_rand). The index_dict.json files contain information on how to read the vectorized models.

For more information on the zoos and code to access and use the zoos, please see www.modelzoos.cc.

Global Fashion Retail Analytics Dataset

📊 Dataset Overview

This synthetic dataset simulates two years of transactional data for a multinational fashion retailer, featuring:
- 📈 4+ million sales records
- 🏪 35 stores across 7 countries:
🇺🇸 United States | 🇨🇳 China | 🇩🇪 Germany | 🇬🇧 United Kingdom | 🇫🇷 France | 🇪🇸 Spain | 🇵🇹 Portugal

Currencies Covered: Each transaction includes detailed currency information, covering multiple currencies:
💵 USD (United States) | 💶 EUR (Eurozone) | 💴 CNY (China) | 💷 GBP (United Kingdom)

Designed for Detailed and Multifaceted Analysis

🌐 Geographic Sales Comparison
Gain insights into how sales performance varies between regions and countries, and identify trends that drive success in different markets.

👥 Analyze Staffing and Performance
Evaluate store staffing ratios and analyze the impact of employee performance on store success.

🛍️ Customer Behavior and Segmentation
Understand regional customer preferences, analyze demographic factors such as age and occupation, and segment customers based on their purchasing habits.

💱 Multi-Currency Analysis
Explore how transactions in different currencies (USD, EUR, CNY, GBP) are handled, analyze currency exchange effects, and compare sales across regions using multiple currencies.

👗 Product Trends
Assess how product categories (e.g., Feminine, Masculine, Children) and specific product attributes (size, color) perform across different regions.

🎯 Pricing and Discount Analysis
Study how different pricing models and discounts affect sales and customer decisions across diverse geographies.

📊 Advanced Cross-Country & Currency Analysis
Conduct complex, multi-dimensional analytics that interconnect countries, currencies, and sales data, identifying hidden correlations between economic factors, regional demand, and financial performance.

Synthetic Data Advantages

Generated using algorithms, it simulates real-world retail dynamics while ensuring privacy.

• Privacy-Safe: All customer and employee data is artificially generated to ensure privacy and compliance with data protection regulations. Personal details, such as emails and phone numbers, are anonymized.
• Scalable Patterns: The data replicates real-world retail dynamics, ensuring scalability of patterns for testing algorithms and analytics models.
• Controlled Complexity: The dataset introduces intentional complexities (e.g., missing job titles, inconsistent phone number formats) to offer a more realistic and challenging exploration experience for exploratory data analysis.
• Customizable for Various Use Cases: Whether you're performing sales forecasting, employee performance analysis, or customer segmentation, this dataset offers a flexible foundation for diverse analytical tasks.

This dataset is an ideal resource for retail analysts, data scientists, and business intelligence professionals aiming to explore multinational retail data, optimize operations, and uncover new insights into customer behavior, sales trends, and employee efficiency.

Here are a few use cases for this project:

1. E-Commerce Retail: The 'labeling' computer vision model can be used to categorize and sort products in online clothing stores. Customers can search for specific clothing items based on type, style, or other features, enhancing the shopping experience and making it easier for retailers to manage their online inventory.

2. Fashion Trend Analysis: Analyzing patterns in clothing choices across different regions or demographics using this model can provide insights into evolving fashion trends. Such information can be crucial for fashion designers, retail stores, and marketers, among others.

3. Personal Wardrobe Organizer: An application could use the model to help individuals organize their wardrobe. It could group clothes by type, style, etc., making it easier for users to find what they want to wear and keep track of their clothes.

4. Lost and Found: It could be used in a lost and found system, categorizing and identifying lost property based on attire type and other characteristics. This makes it easier to match and retrieve lost clothing items.

5. Augmented Reality Shopping: This computer vision model could be integrated into an AR-based shopping app. Virtual mannequins could wear clothes detected by the model, enabling customers to visualize how the clothing might look on them before making a purchase.

This dataset contains images (scenes) containing fashion products, which are labeled with bounding boxes and links to the corresponding products.

Metadata includes

• product IDs

• bounding boxes

Basic Statistics:

• Scenes: 47,739

• Products: 38,111

• Scene-Product Pairs: 93,274

In this study, 403 Chinese consumers generalizable to the broader population were surveyed on their motivations to shop for fashion apparel in both high street and e-commerce environments. Statistical analysis was undertaken through multiple T-Tests and MANOVA with the assistance of SPSS and G*Power.

To increase the profits of international brands, this paper presents the motivations of Chinese consumers to engage in fashion retail, building upon established theory in hedonic and utilitarian motivations. With China set to capture over 24% of the $212 billion fashion market, international brands need to understand the unique motivations of Chinese consumers in order to capitalise on the market. However, the motivations of Chinese people to engage in fashion retail are as yet undefined, limiting the ability for international fashion retailers to operate with prosperity in the Chinese market.

Imports - Nontextile Apparel & Hh. Goods (Census Basis) in the United States increased to 771.43 USD Million in February from 731.09 USD Million in January of 2024. This dataset includes a chart with historical data for the United States Imports of Nontextile Apparel & Hh. Goods.

Exports - Apparel, Footwear & Household (Census Basis) in the United States decreased to 1007.19 USD Million in February from 1012.61 USD Million in January of 2024. This dataset includes a chart with historical data for the United States Exports of Apparel, Footwear & Household.

We'll tailor a Chictopia dataset to meet your unique needs, encompassing outfit details, user engagement metrics, fashion trends, demographic data of users, social media interactions, style ratings, and other pertinent metrics.

Leverage our Chictopia datasets for diverse applications to bolster strategic planning and market analysis. Scrutinizing these datasets enables organizations to grasp fashion preferences and industry trends, facilitating nuanced product selection and marketing initiatives. Customize your access to the entire dataset or specific subsets as per your business requisites.

Popular use cases involve pricing optimization by creating dynamic pricing models based on competitor comparisons, identifying gaps in product inventory and trending consumer preferences, and optimizing market strategy by analyzing key fashion trends and customer preferences to enhance decision-making in the fashion industry.

Imports - Apparel, Footwear & Hh. Goods, Cotton (Census Basis) in the United States increased to 3425.20 USD Million in February from 3146.66 USD Million in January of 2024. This dataset includes a chart with historical data for the United States Imports of Apparel, Footwear & Hh. Goods, Cotton.

This data set was collected through online self-administrative survey. Sample size was a total of 223 hijabistas, and filtered into 188 usable responses (84.3% per cent response rate).
The survey had three main sections: 1. Introduction, instructions, 2. Demographics (age, material status, occupation, and ethnicity) of the respondents, along with two filter questions to define their religious obligations in daily life and their fashion consciousness; 3. Survey items that measure seven main variables defined as “product quality”, “design quality”, “self-identity”, “social identity”, “product satisfaction”, “brand satisfaction”, and premium price.

A comprehensive dataset providing insights into the fashion industry, including market size, employment statistics, trends, and analysis on the apparel sector for 2025.

Graph and download economic data for Retail Sales: Clothing and Clothing Accessory Stores (MRTSSM448USN) from Jan 1992 to Apr 2025 about apparel, retail trade, sales, retail, and USA.

We'll tailor a Nike dataset to meet your unique needs, encompassing product details, user engagement metrics, purchasing trends, demographic data of customers, social media interactions, product ratings, and other pertinent metrics.

Leverage our Nike datasets for diverse applications to bolster strategic planning and market analysis. Scrutinizing these datasets enables organizations to grasp consumer preferences and industry trends, facilitating nuanced product selection and marketing initiatives. Customize your access to the entire dataset or specific subsets as per your business requisites.

Popular use cases involve pricing optimization by creating dynamic pricing models based on competitor comparisons, identifying gaps in product inventory and trending consumer preferences, and optimizing market strategy by analyzing key sportswear trends and customer preferences to enhance decision-making in the athletic apparel industry.

Imports - Apparel, Footwear & Hh. Goods, Wool (Census Basis) in the United States increased to 256.76 USD Million in February from 246.20 USD Million in January of 2024. This dataset includes a chart with historical data for the United States Imports of Apparel, Footwear & Hh. Goods, Wool.

These datasets contain peer-to-peer trades from various recommendation platforms.

Metadata includes

• peer-to-peer trades

• have and want lists

• image data (tradesy)

Exports - Military Apparel & Footwear (Census Basis) in the United States decreased to 2.66 USD Million in February from 8.11 USD Million in January of 2024. This dataset includes a chart with historical data for the United States Exports of Military Apparel & Footwear.

This is the dataset for a research project called the Virtual Shoe Salon. The Virtual Shoe Salon is a project that focuses on young people, aged 18-24 years, and their sense of self as it is communicated and experienced through their footwear choices. The participants were asked to take photographs of their shoes and write a short narrative response to what the shoes mean to them and how the shoes make them feel as an individual. This dataset includes photographs, along with each individual's written narratives and demographic information.

TheLook is a fictitious eCommerce clothing site developed by the Looker team. The dataset contains information about customers, products, orders, logistics, web events and digital marketing campaigns. The contents of this dataset are synthetic, and are provided to industry practitioners for the purpose of product discovery, testing, and evaluation. This public dataset is hosted in Google BigQuery and is included in BigQuery's 1TB/mo of free tier processing. This means that each user receives 1TB of free BigQuery processing every month, which can be used to run queries on this public dataset. Watch this short video to learn how to get started quickly using BigQuery to access public datasets.What is BigQuery .

This table contains 50688 series, with data for years 2012 - 2012 (not all combinations necessarily have data for all years). This table contains data described by the following dimensions (Not all combinations are available): Geography (11 items: Canada; Atlantic; Quebec; Ontario; ...) Aboriginal identity (6 items: Total, Aboriginal identity; First Nations (North American Indian); First Nations (North American Indian), Registered or Treaty Indian; First Nations (North American Indian), not a Registered or Treaty Indian; ...) Age group (4 items: Total, 15 years and over; 15 to 24 years; 25 to 54 years; 55 years and over) Sex (3 items: Both sexes; Male; Female) Making handcrafted goods (16 items: Total, made clothing or footwear in the last year; Made clothing or footwear in the last year; Made clothing or footwear in the last year for pleasure or leisure; Made clothing or footwear in the last year for own or family's use or to supplement income; ...) Statistics (4 items: Number of persons; Percent; Low 95% confidence interval; High 95% confidence interval).

This version (Version 6) is the dataset associated with the article, Duncanson, K.A.; Thwaites, S.; Booth, D.; Hanly, G.; Robertson, W.S.P.; Abbasnejad, E.; Thewlis, D. Deep Metric Learning for Scalable Gait-Based Person Re-Identification Using Force Platform Data. Sensors 2023, 23, 3392. https://doi.org/10.3390/s23073392. Please see Version 3 for the dataset and description relevant to the article, Duncanson, Kayne; Thwaites, Simon; Booth, David; Abbasnejad, Ehsan; Robertson, William; Thewlis, Dominic (2021): The Most Discriminant Components of Force Platform Data for Gait Based Person Re-identification. TechRxiv. Preprint. https://doi.org/10.36227/techrxiv.16683229.v1.Dataset overviewThis dataset was acquired for research on the use of gait as a (soft) biometric for person re-identification (re-ID)/recognition; however, it may be used to answer a variety of research questions. It is one of the largest and most complex force platform datasets purpose built for person re-ID. It contains 5327 walking trials from 184 healthy participants, with inter- and intra-individual variation in clothing, footwear, and walking speed, as well as inter-individual variation in time between trials (data was collected over two sessions separated by 3-14 days depending on the individual). The dataset was generated through a repeated measures experiment (approved by the Human Research Ethics Committee - approval No. H-2018-009) conducted at The University of Adelaide gait analysis laboratory.Experimental protocolAt the start of each session, age, sex, mass, height, and footwear type were recorded, as participants wore personal clothing and footwear. Footwear was also photographed for future reference. Next, participants walked in one direction along the length of the laboratory (≈10m) five times at three self-selected speeds: preferred, slower than preferred, and faster than preferred. Two in-ground OPT400600-HP force platforms (Advanced Mechanical Technology Inc., USA) in the center of the laboratory measured GRFs and GRMs during left and right footsteps. These measures, along with calculated COP coordinates, were acquired through Vicon Nexus (Vicon Motion Systems Ltd, UK) at 2000 Hz. Of note, all trials in this dataset were complete foot contacts; that is, each foot contacted completely within the area of each force platform, as identified from video footage.User guideThe code repository to implement this dataset can be found at GitHub - kayneduncanson1/ForceID-Study-1: Repository for the article, 'Deep Metric Learning for Scalable Gait Based Person Re-identification Using Force Platform Data'. The dataset is organised into separate spreadsheets that each contain all samples of a particular component from a given force platform (named as component_platform). Both raw and processed ('pro') versions are available. Within each of the data spreadsheets is accessory information about each trial in the first five columns, followed by the data in column six onward. Within the metadata spreadsheet are the ID numbers and their associated demographics. The ID numbers in the first column are from the private version of the dataset that was implemented for the manuscript. These can be cross-referenced with the ID numbers generated for the public dataset in the code repository. This means that the challenging IDs listed in Supplemental Material - Table VII (in the article) can be located in this dataset.