Motion-o-MCoT: 3,168 Examples with Motion Reasoning and Keyframes

A subset of the STGR dataset, Motion-o-MCoT contains 3,168 training examples where the reasoning process involves motion. The dataset includes associated keyframe images stored as JPEGs in sharded directories. It was created by author bishoygaloaa and was last updated on March 29, 2026.

Use Cases

• Training models for visual reasoning tasks based on motion sequences described in the reasoning process.
• Developing multimodal systems that link textual reasoning steps to corresponding image keyframes.
• Analyzing the relationship between motion concepts in language and their visual representations.
• Benchmarking AI performance on tasks requiring joint understanding of logic and visual dynamics.

Strengths

• Contains 3,168 examples for training, providing a defined scale.
• Includes motion reasoning annotations, a specific feature mentioned in the description.
• Keyframe images are provided, enabling multimodal analysis.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown for the full dataset, which may limit suitability assessment.

Provenance

Source

Subset of the STGR dataset, specifically STR_plm_rdcap rows.

Collection Method

Rows were selected where the reasoning process contains motion.

Time Range

null

Freshness

Last updated 2026-03-29 18:13:45; freshness should be verified.

Geography

null