Howto-Interlink7M
About 1 million original YouTube videos, 85TB, collected in August 2024, https://huggingface.co/datasets/Awiny/Howto-Interlink7M
A large-scale teaching video dataset containing 7 million interrelated tutorial video clips. It connects relevant steps from different tutorials through procedural knowledge, supporting program understanding, step recognition, and cross-video reasoning research.
Large-scale cross-video instructional data, promoting program understanding and multimodal research
Contains over 7 million instructional video clips, making it one of the largest tutorial video datasets, providing a solid foundation for large-scale multimodal research.
A unique cross-video connection mechanism based on program similarity that links semantically related steps from different tutorials, supporting cross-video knowledge transfer.
Each video clip has fine-grained step description annotations, accurately depicting each operational step in the instructional process, providing rich semantic information.
Rich procedural structures capture complete operational processes, organizing scattered instructional steps into an orderly knowledge system, aiding process reasoning research.
Covers instructional content in various life domains such as cooking, crafts, maintenance, and gardening, ensuring the dataset's broad applicability and diversity.
Each instructional step is annotated with precise time boundaries, supporting accurate localization of video clips and temporal modeling research.
From basic research to practical applications, unleash the potential of instructional video data
Train models to understand multi-step processes, learning the sequences and causal relationships in instructional videos
Predict the next action in the instructional sequence, building intelligent models that can foresee the flow of processes
Find related instructional steps across different tutorial videos, achieving cross-video knowledge discovery
Build AI assistants that can guide users through operational processes, providing intelligent instructional guidance
Quickly access the dataset through the Ace Data Cloud API
import requests
# Set your API token
API_TOKEN = "your_api_token_here"
# Request dataset information
response = requests.get(
"https://api.acedata.cloud/datasets/howto-interlink7m",
headers={
"Authorization": f"Bearer {API_TOKEN}",
"Accept": "application/json"
}
)
# Parse response
data = response.json()
print(f"Dataset: {data['name']}")
print(f"Total clips: {data['total_clips']}")
print(f"Domains: {data['domains']}")
# Access video clip details
for clip in data["clips"][:5]:
print(f" Step: {clip['step_description']}")
print(f" Start: {clip['start_time']}s")
print(f" End: {clip['end_time']}s")
print(f" Links: {len(clip['interlinks'])} related clips")
print()
From registration to use, you can start accessing the dataset in just a few minutes
Go to platform.acedata.cloud to register your Ace Data Cloud account and quickly complete developer authentication.
Create an API key in the console to obtain the credentials to access the HowTo-InterLink7M dataset.
Use your API Key to call the dataset interface and start accessing 7 million teaching video clips and cross-video related data.
7 million teaching video clips with a unique cross-video association structure. Whether you are researching program understanding, video retrieval, or teaching AI, this dataset is an ideal choice.