AMVOS: Additive Manufacturing Video Object Segmentation Dataset

900 individually annotated video frames across 90 segments from five additive manufacturing processes. Data was collected at Oak Ridge National Laboratory's Manufacturing Demonstration Facility and is structured for direct integration into video object segmentation model pipelines.

Use Cases

• Fine-tuning video object segmentation foundation models based on labeled AM process videos.
• Benchmarking zero-shot performance of VOS models on domain-specific manufacturing data.
• Training models for domain adaptation from general VOS benchmarks to the AM domain.
• Analyzing the dynamics of object classes like Melt Pool or Feed Wire during manufacturing.

Strengths

• 900 individually annotated frames provide pixel-level instance labels.
• Data follows the directory structure of established VOS benchmarks (DAVIS, YouTube-VOS, MOSE) for easy integration.
• Covers five distinct additive manufacturing processes (e.g., LHW-DED, TIG-WAAM).

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

Oak Ridge National Laboratory's Manufacturing Demonstration Facility.

Collection Method

Likely collected via video recording of manufacturing processes with subsequent manual or semi-automated annotation.

Freshness

Last updated 2026-05-15 21:24:59; freshness should be verified.