RSK World - Action Recognition Dataset | Video Classification | Video ML | Action AI | RSK World

Action Recognition Video Classification Machine Learning 1,500+ Videos Python 3D CNN Ready

This project features a comprehensive Action Recognition dataset designed for professional video action classification, video understanding, and action analysis applications. The dataset includes 1,500+ labeled video files across 12+ action classes (walking, running, jumping, dancing, exercising, yoga, stretching, boxing, etc.), preprocessed video sequences, frame-level annotations, and Python scripts for model training. Includes powerful Python scripts: train_model.py for 3D CNN model training (C3D, I3D, R3D), loader.py for dataset loading, augmentation.py for video augmentation, api_server.py for REST API, realtime_predictor.py for webcam prediction, and benchmark.py for model evaluation. The package includes interactive demo website, REST API server, real-time webcam predictor, analytics dashboard, comprehensive README.md, and MIT License. Perfect for data scientists, researchers, students, and developers working on action recognition, video classification, video understanding, and video analysis projects.

If you find this Action Recognition Dataset useful, you can support with a small contribution.

Secure Fast Trusted

Pay via UPI QR

Scan or tap an amount to auto-generate

₹

Open UPI app

GPay PhonePe Paytm

Download Free Source Code

Dataset Overview

Complete action recognition dataset with 1,500+ labeled video files across 12+ action classes for video AI and machine learning.

1,500+ labeled video files
12+ action classes (walking, running, jumping, dancing, exercising, yoga, stretching, boxing, etc.)
3-5 seconds per video clip
MP4 and AVI formats
640x480 resolution
30 FPS frame rate
Pre-labeled with action tags
Balanced class distribution
Real human action videos
Training/validation/test split
Frame-level annotations available
Perfect for action recognition & ML training

Dataset Structure & Files

Well-organized folder structure with video files organized by action class, annotations, and metadata.

sample_data/ - Video files organized by action class (MP4/AVI)
annotations/ - JSON annotation files (train, val, test)
sample_data/train/ - Training videos by action class
sample_data/val/ - Validation videos by action class
sample_data/test/ - Test videos by action class
Python scripts - train_model.py, loader.py, augmentation.py
API & Demo - api_server.py, realtime_predictor.py, demo.html
Consistent naming convention
Easy to load with PyTorch DataLoader
PyTorch/TensorFlow ready format
Frame-level annotations preprocessed

Machine Learning Training

Complete training pipeline with support for 3D CNNs, video classification models, and deep learning frameworks.

3D CNN models (C3D, I3D, R3D)
SlowFast networks
Video Transformer support
Two-Stream networks
PyTorch support
TensorFlow/Keras compatibility
Video frame extraction
Temporal feature extraction
Batch processing support
Model checkpointing
Performance metrics report
Hyperparameter tuning
Model export & persistence

Multiple File Formats

Dataset available in multiple formats for maximum compatibility with different video processing tools and frameworks.

MP4 format (compressed video)
AVI format (uncompressed video)
JSON format for annotations and metadata
OpenCV compatible
NumPy array format for frames
Easy format conversion
640x480 resolution
30 FPS frame rate
Standard video formats
Compatible with all video libraries
PyTorch DataLoader ready

Analysis & Visualization

Comprehensive analysis tools with visualization capabilities and interactive action explorer.

Interactive Action Explorer
Class distribution charts
Video frame visualization
Motion heatmap visualization
Temporal feature plots
Feature correlation analysis
Performance benchmarking
Model comparison tools
HTML report generation
Export visualization images
Analytics Dashboard

Compatible Frameworks

Works with all major video AI and machine learning frameworks out of the box.

PyTorch deep learning
TensorFlow/Keras
OpenCV video processing
NumPy numerical computing
scikit-learn ML library
pandas data manipulation
matplotlib visualization
Jupyter Notebook support
3D CNNs (C3D, I3D, R3D)
Video Transformers
Neural Network models

What You Get

Complete package with all files needed for professional action recognition projects.

1,500+ labeled video files
Python utility scripts
train_model.py - 3D CNN model training
loader.py - Dataset loading
augmentation.py - Video augmentation
api_server.py - REST API server
realtime_predictor.py - Webcam prediction
benchmark.py - Model benchmarking
Interactive demo website
Video player integration
Video processing pipeline
Complete documentation

Interactive Demo Website

Beautiful demo website with action explorer, live video player, analytics dashboard, and comprehensive guide.

Modern animated design
Interactive Action Explorer
Live Video Player
Analytics Dashboard
Filter by action class
Video frame visualization
Class distribution charts
Performance metrics display
Step-by-step usage guide
Dark theme with gradients
Fully responsive layout

Python Scripts Included

Professional Python scripts for video processing, feature extraction, and model training.

train_model.py - 3D CNN model training (C3D, I3D, R3D)
loader.py - Video dataset loading & preprocessing
augmentation.py - Video augmentation pipeline
api_server.py - REST API for video predictions
realtime_predictor.py - Real-time webcam prediction
benchmark.py - Model performance benchmarking
visualize_dataset.py - Dataset visualization
process_downloaded.py - Video processing utilities
Batch processing support
Video feature extraction pipeline
Model training utilities
Complete code examples

Dataset Features

Comprehensive action recognition dataset with 12+ action classes and various video characteristics.

12+ action classes - walking, running, jumping, dancing, exercising, yoga, stretching, boxing, etc.
Diverse action styles - Robust training
Real human videos - Generalizable models
Balanced class distribution
3-5 second clips - Consistent format
High-quality video recordings
Accurate action labels
Frame-level annotations
Training/validation/test split included
Easy to extend dataset
Total: 1,500+ video files