Oak Ridge National Laboratory's Manufacturing Demonstration Facility collected 90 video segments comprising 900 annotated frames from four additive manufacturing processes. The dataset includes pixel-level instance annotations for object classes like Melt Pool, Feed Wire, Nozzle, and Material across five specific AM techniques. It was published by Calvin Wetzel on the Harvard Dataverse platform.
Use Cases
- Fine-tuning foundation models for video object segmentation based on labeled AM process videos.
- Benchmarking zero-shot performance of VOS models on the domain of additive manufacturing.
- Training models for domain adaptation from general VOS benchmarks to specialized industrial video.
- Developing models to segment and track objects like melt pools and feed wires in manufacturing videos.
Strengths
- Contains 900 individually annotated frames across 90 video segments.
- Covers five distinct additive manufacturing processes: LHW-DED, TIG-WAAM, PAW, visPolymer, and irPolymer.
- Follows the directory structure of established VOS benchmarks (DAVIS, YouTube-VOS, MOSE) for direct integration.
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
- Data was collected from four additive manufacturing processes.
- Freshness
- Last updated 2026-05-14 01:38:45; freshness should be verified.