googlecreativelab/quickdraw-datasetArchived
Quickdraw Dataset
Acest proiect este un set de date la scară largă cu schițe desenate manual, oferind milioane de desene vectoriale și bitmap-uri cu marcaj temporal pentru antrenarea modelelor de machine learning. Servește drept corpus de antrenament pentru viziune computerizată și set de date pentru rețele neuronale, constând în schițe umane categorisite, utilizate pentru a dezvolta algoritmi de clasificare și recunoaștere a imaginilor.
Setul de date este disponibil ca un corpus de desene vectoriale care prezintă secvențe de linii și metadate, precum și ca array-uri numpy procesate. Aceste resurse susțin dezvoltarea clasificatorilor de desene și studiul tiparelor de desen uman.
Datele sunt furnizate în mai multe formate, inclusiv date vectoriale brute în JSON delimitat prin linii noi, secvențe vectoriale normalizate și bitmap-uri grayscale. Include capabilități pentru partiționarea bazată pe categorii și scalarea coordonatelor pentru a asigura consistența între diferite eșantioane.
Features
- Computer Vision Datasets - Provides a large-scale collection of categorized human sketches as image and vector data for training recognition models.
- Vector Drawing Dataset Downloads - Provides large collections of timestamped vector drawings across multiple categories in various formats for ML research.
- Coordinate Normalization Utilities - Rescales raw pixel coordinates to a consistent range to remove variance caused by different drawing screen sizes.
- Hand-Drawn Sketch Datasets - Offers millions of timestamped vector drawings and bitmaps specifically curated for training machine learning models.
- Neural Network Training Datasets - Supplies processed numpy and JSON data formatted for use in recurrent and convolutional neural networks.
- RNN Dataset Access - Provides a method to retrieve compressed numpy files formatted specifically for training recurrent neural networks.
- Sketch-Based Machine Learning - Developing neural networks that process vector stroke data or bitmap images to analyze human drawing patterns.
- Sketch Classifiers - Provides data for training models to recognize and classify hand-drawn sketches and symbols across hundreds of categories.
- Raw Data Retrieval - Retrieves unsimplified drawings containing pixel coordinates, timing information, and metadata in JSON format.
- Vector Data Processing - Provides a method to retrieve rescaled drawings processed to remove timing information for consistent input.
- Vector Drawing Corpora - Provides a massive archive of stroke-by-stroke drawing sequences and metadata for stroke generation research.
- 2D Vector Representations - Records drawings as sequences of coordinates and timestamps to preserve the temporal order of human sketching.
- Numpy Bitmap Access - Retrieves drawings rendered as grayscale bitmaps in a numpy format for use with image-based neural networks.
- Preprocessed ML Bitmaps - Retrieves simplified drawing data rendered as grayscale bitmaps for use in image-based classification models.
- Stroke Data Serializations - Retrieves compressed data formatted for recurrent neural networks to study stroke-by-stroke drawing generation.
- CNN Input Rasterization - Converts coordinate sequences into grayscale grids for compatibility with convolutional neural network input layers.
- Drawing Classifiers - Supports the development and evaluation of machine learning models designed to classify hand-drawn sketches.
- Data Preprocessing for Modeling - Converts raw drawing vectors into normalized formats or grayscale bitmaps for use in recurrent and convolutional networks.
- Pre-computed Dataset Archives - Distributes processed data in compressed numpy formats to reduce download time and memory overhead for researchers.
- Newline-Delimited JSON Streams - Stores large datasets as separate JSON objects per line to allow efficient streaming and partial file reading.
- Visual Cognition Research - Supports studying how people from different countries visualize common objects through timestamped vector drawings.
- Game Datasets - Large-scale collection of human-drawn sketches for machine learning.
