InfoBayAI/stem-videos

Dataset Description:

This dataset is a large-scale collection of STEM educational video data, containing 100,000 hours of video content, designed to support the development and training of advanced video understanding, multimodal AI, vision-language models (VLMs), educational AI systems, video captioning, video reasoning, and large-scale machine learning applications.

Additionally, this dataset can be used in pipelines for Supervised Fine-Tuning (SFT) and Reinforcement Learning with Human Feedback (RLHF) workflows, improving model performance in scientific reasoning, video-based learning, and multimodal problem-solving tasks.

Video Refining Process To ensure enterprise-grade quality and usability, the dataset undergoes a comprehensive 8-step video refining and processing pipeline before final delivery:

1. Duplicate Asset Elimination — Removal of duplicate or repeated video assets to maintain dataset uniqueness, consistency, and high-quality training data.
2. Vertical Format Validation — Verification that all video content meets vertical format compliance standards for mobile-first AI training requirements.
3. AV1 Codec Audit — Standardization of video codec to H.264 to ensure broad compatibility and consistent playback across platforms and models.
4. Intelligent Text Recognition — Automated ITR processing to detect, extract, and handle on-screen text within video frames for cleaner training data.
5. Animated Content Classification — Type tagging and classification of animated vs. real-world content to enable accurate dataset segmentation.
6. AI-Synthetic Media Detection — Authenticity filtering to identify and flag AI-generated or synthetic video content, ensuring real-world data integrity.
7. Brand Logo Intelligence — BLI detection to identify, track, and manage brand logos within video frames for compliance and neutrality.
8. Watermark Integrity Analysis — Detection and removal of watermarks to deliver clean, unobstructed video output ready for AI model training.

This processing pipeline ensures that the dataset is clean, scalable, production-ready, and optimized for video AI, computer vision, multimodal AI, SFT, and RLHF workflows.

Dataset Specification

Duration: 100,000 Hours
Modality: Video
Domain: STEM Education
Type: Educational Video Content
Data Nature: Real-world and curated content

Key Use Cases

Video Understanding
Vision-Language Model (VLM) Training
Educational AI Systems
Video Captioning
Video Question Answering (Video QA)
Multimodal Learning
Video Retrieval
Action and Activity Recognition
Visual Reasoning
AI-Powered Learning Platforms
Large Language Models with Video Understanding

Value of This Dataset

-Enables advanced STEM learning through video-based AI models
-Improves scientific reasoning and conceptual understanding
-Supports multimodal AI systems (vision + audio + text)
-Helps build AI-powered tutoring and educational platforms
-Enhances performance in video-based question answering and summarization
-Strengthens LLMs with multimodal STEM reasoning capabilities

Basic Json Schema

{
  "file_name": "string",
  "format": "string",
  "duration_seconds": "float",
  "width": "int32",
  "height": "int32",
  "fps": "float",
  "video_codec": "string",
  "audio_codec": "string",
  "bitrate": "int64",
  "file_size_bytes": "int64"
}

Data Creation

Procured through formal agreements and generated in the ordinary course of business.

Considerations

This dataset is provided for research and educational purposes only. It contains only sample data. For access to the full dataset and enterprise licensing options, please visit our website InfoBay.AI or contact us directly.

-Ph: (91) 8303174762
-Email: datareq@infobay.ai