Sama Material Centric Video Dataset

500 synthetic video samples with dense per-frame, per-material pixel-level segmentation annotations, introduced alongside the SAMa research paper. The dataset, created by Voxel51, is split into 450 training and 50 test samples for fine-tuning video object-selection models. It was last updated on March 17, 2026.

Use Cases

• Fine-tuning video object-selection models for material selection based on dense pixel-level annotations.
• Training models for per-frame, per-material segmentation in synthetic video scenes.
• Evaluating the performance of segmentation models on a dedicated test set of 50 samples.
• Researching material-aware 3D selection and segmentation techniques in computer vision.

Strengths

• 500 total samples with a defined 450/50 train/test split.
• Dense per-frame, per-material pixel-level segmentation annotations.
• Designed specifically for the material selection task in video.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data is synthetic, which may not fully represent real-world video characteristics.

Provenance

Source

Voxel51

Collection Method

Synthetic video dataset created for the SAMa research project.

Time Range

null

Freshness

Last updated 2026-03-17 19:19:34; freshness should be verified.

Geography

null