Assembly101: Multi-View Videos of Toy Vehicle Assembly and Disassembly

Assembly101 is a procedural activity dataset featuring 4321 videos of people assembling and disassembling 101 "take-apart" toy vehicles. Participants worked without fixed instructions, resulting in natural variations in action ordering, mistakes, and corrections. The dataset was created by cvml-nus and includes simultaneous static and egocentric recordings.

Use Cases

• Training action recognition models based on coarse and fine-grained action annotations
• Developing models for procedural activity understanding based on natural variations in assembly sequences
• Researching multi-view video analysis based on simultaneous static and egocentric recordings
• Studying human error and correction patterns in procedural tasks based on unscripted assembly actions

Strengths

• Contains 4321 videos, providing a substantial collection for training
• Features 101 distinct "take-apart" toy vehicles as assembly subjects
• Includes more than 100K coarse and 1M fine-grained action annotations
• Is the first multi-view action dataset with simultaneous static (8) and egocentric (4) recordings

Limitations

• Column-level documentation is absent; field semantics must be inferred after download
• Row count is unknown, which may limit suitability assessment
• Data may reflect bias inherent to the specific toy vehicles and participant pool used

Provenance

Source

cvml-nus via Hugging Face

Collection Method

Recordings of participants assembling and disassembling toy vehicles without fixed instructions.

Freshness

Last updated 2026-05-22 04:03:00; freshness should be verified