AI Training Datasets

Dataset · Object detection

flame_smoke_detection_box_not_precise

Fire Viewer

A prepared two-class detection dataset with 38,965 unique native images and 50,591 bounding boxes. It combines the retained FireViewer detection source snapshot with the published negative corpus, with reproducible grouped splits and a COCO materializer. The legacy repository identifier is retained; this card describes the v2.5 release recorded in manifest.json. Train also includes 504 negative images from the detection source; the negative columns above identify only the two imported negative subsets. Native-negative difficulty is unrated; the JPEGs retain their historical candidate designation. COCO category IDs are 1: smokevisible and 2: flamevisible. Map them explicitly to internal…

This dataset is continuously seeded from OpenStreetMap, matched to Google Place IDs, harvested for temporal store photos, and enriched with zero-shot computer vision using Hugging Face Hub native pipelines. The dataset is partitioned into three relational subsets loadable via Hugging Face datasets: Contains every verified physical retail location with multi-authority cross-references: - businessid: Deterministic unique identifier - name, brand, category - adminlevel0 (Country) -> adminlevel1 (State) -> adminlevel2 (City/LGA) -> adminlevel3 (Suburb) - osmid & osmurl: Canonical OpenStreetMap reference - googleplaceid & googlemapsurl: Google Maps live location - latitude, longitude Contains…

Dataset · Audio to audio

speech-to-speech-collections

Meghana Bommadi

1444 train / 23 validation calls from a production collections voice agent (Hindi/Hinglish), processed into the same schema as MeghanaKap/speech-to-speech-data. sides can be modelled without speaker separation. Real customers are heard in these recordings, so access is gated and reviewed. Kept: the customer's first name (the model must learn to say it), the amounts discussed, and the last 4 digits of the loan the agent reads aloud. Removed: full loan/account numbers, phone numbers (muted in the audio, tagged in the text, and stripped from the word timings), conversation and client ids, the source recording URL, and any digit run of 5 or more in the reference script. slotdict / systemprompt…

Raw text from 4 sources (light cleaning: NFKC, diacritics/tatweel, URLs; exact de-duplication; no content classification), chat conversations, and token shards. Random sample of AdaMLLab/AraMix / sentencededuped: 50,000,000,000 tokens (counted with this project's tokenizer), no classification and no cleaning (only exact de-duplication). Stored as token shards in tokensv3128k/ (no raw-text copy). - wikimedia/wikipedia - HuggingFaceFW/fineweb-2 - Gabrui/multilingualTinyStories - Jr23xd23/rightnow-arabic-llm-corpus - 2A2I/Arabic-OpenHermes-2.5 - ClusterlabAi/InstAr-500k - FreedomIntelligence/Evol-Instruct-Arabic-GPT4 - FreedomIntelligence/alpaca-gpt4-arabic - MBZUAI/Bactrian-X - arbml/CIDAR…

Hidden-state activations from google/gemma-3-27b-it, recorded while it lied or told the truth and was then asked whether it had lied. Part of a SPAR replication of Anthropic's post, built on its released dataset Each tensor is labelled with independently checked ground truth (did gemma actually lie on this row?), not the dataset's own didlie column, which is unusable for gemma. unanswerable, capturesecret, alibi, and the MASK belief subtypes maskknownfacts, maskstatistics, maskcontinuations, maskdisinformation (their first-run state is the pressured prompt: firstrun/mask / pressuredpredecision.pt). After gemma's answer, it is asked whether it lied, with A = yes and B = no: - Shape [63…

One row per player, game, team, and statistic group. Regular-season games at AA, High-A, and Single-A, subject to configured coverage. Source: MLB Stats API. Data is extracted from box-score stats, never seasonStats. Missing statistics are null. pitchingIPstr uses baseball notation; use pitchingouts for arithmetic. Files are grouped by season, historical league ID, and team ID. catalog.json maps IDs to names and table paths. The train split is a loading convention; no training/test split has been applied. state/index.json records season completion; partially backfilled seasons can be visible. state/SEASON.json records the games successfully stored. Current-season games are refreshed…

Dataset · Image to text

IndicOCR

Faizan Iqbal

Large-scale multilingual OCR and document dataset across 23 Pan-Indic languages and 12 writing systems. IndicOCR (IndicPixel) is a large-scale multilingual Optical Character Recognition (OCR) and document dataset covering the South Asian linguistic landscape. The dataset provides dense document coverage across 23 official and literary languages representing 12 distinct writing systems. 4-Tier Document Hierarchy: Encompasses complete structural representation from isolated tokens to complex multi-column documents: Each sample is packaged inside standardized WebDataset.tar archives containing paired lossless image files (.webp) and comprehensive metadata descriptors (.json): samplekey…

最后更新 (last updated): 2026-10-18 11:32 CST (UTC+8) 总进度 (overall): 84490/167353 (50.5%) 文件 (files), 已上传 (uploaded) 253.38 GB；82 个产品 (products) 分类说明见下文 Classification and folder layout。类型 (type): obs = 原始/L1–L2 观测；retrieval product = L3/L4 卫星反演产品；reanalysis/model = 再分析/模式；derived index/climatology = 派生指数/气候态。 Unmodified mirror of public Arctic and Antarctic sea-ice products (satellite observations and retrievals, reanalyses/model products, derived indices and climatologies), collected for validating CICE sea-ice model runs. Files keep their original names; every file is listed in manifest.csv (size, MD5, checksum source, original URL, classification columns, licence, verified on Hugging Face /…

Dataset

amc-rues

Carlos Tuiran

El registro empresarial de Colombia (RUES), convertido de 9,4 millones de matrículas en texto plano en una fila por empresa, con su vigencia, su sector y su organización jurídica explícitos. Construido y verificado por el Solo se publica lo que pasa la verificación. Cada archivo se comparó con la base de datos de la que salió, y una verificación fallida no publica nada. En el navegador, sin instalar nada: la consulta SQL de AMC. O con DuckDB, que solo descarga las partes que la consulta necesita: companies tiene una fila por persona jurídica o comerciante. Los demás son los no es una tasa de supervivencia), el embudo de la lista objetivo, la cobertura (cada población excluida de una cifra…

Dataset · Tabular classification

RaceData

Tracing Insights

Multi-table Formula 1 dataset. Each table is exposed as a separate configuration so that every table keeps its own column schema.

English | 中文 Physically-annotated indoor scene USD assets derived from 14,169 real interior design schemes on the sunvage floorplan platform — every furniture instance carries collision shapes, mass, friction and restitution, and opens directly in USD physics simulators. sunvage-vertens3d-simhome is a collection of physically-annotated indoor scene USD assets derived from real interior design schemes in the sunvage (三维家) floorplan library. It provides both whole-home multi-space scenes and individual single-space scenes. Every physical instance carries collision shapes and mass properties, and the scenes can be loaded directly into USD physics simulators such as Isaac Sim / Isaac Lab for…

Team nau shunya do sau das: Rudr Pratap Singh, Uday Bhatia, Swayam Singh, Antriksh Arya. The archive holds our code, experiment records, models and final submission for the Business Entity Resolution challenge: match Source 2/3 business records to Source 1 entities, scored by macro F0.5. The final submission's expected public score is 0.98564. See finalsubmission/FINALB20260927/Documentationtemplate.md for the method and workspace/CONTEXTHANDOFF.md for the full project history. The competition dataset itself is included under workspace/data/raw/; it belongs to the challenge organisers and remains subject to their terms.