4DThinker: Training Data for Dynamic Latent Mental Imagery in VLMs

Training data for the 4DThinker framework, which enables Vision Language Models to 'think with 4D' through dynamic latent mental imagery. The dataset includes approximately 38,000 samples for DIFT training and 37,000 samples for 4DRL training, built upon SpatialVID and DSR_Suite-Data. It was authored by jankin123 and last updated on May 11, 2026.

Use Cases

• Train Vision Language Models on dynamic scene representations based on the described dynamic latent mental imagery framework.
• Develop models for spatiotemporal reasoning based on video frames and mask overlays mentioned in the data structure.
• Benchmark multimodal learning approaches based on the combined DIFT and 4DRL training samples.
• Fine-tune models for tasks requiring integration of visual and textual data over time based on the 4DThinker framework.

Strengths

• Contains approximately 38,000 DIFT training samples.
• Contains approximately 37,000 4DRL training samples.
• Includes processed video frames and mask overlays, suggesting structured multimodal content.

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, which may limit suitability assessment.

Provenance

Source

huggingface

Collection Method

Built upon SpatialVID and DSR_Suite-Data.

Freshness

Last updated 2026-05-11 21:55:06; freshness should be verified.