GSA_volc

GSA volc – GSA Embodied Perception Training Dataset Large-scale Grounding-Spatial-Affordance (GSA) training data for embodied perception Teacher models. Designed for plug-and-play usage with ms-swift. Data Summary Metric Count Total Images ~580K Total Annotations 3,267,527 SFT Samples (default) 1,000,000 GRPO Samples (default) 250,526 SFT Samples (bbox grounding 1m) 1,000,000 GRPO Samples (bbox grounding 1m) 250,000 Capability Dimensions 6… See the full description on the dataset page: volc.

Source: Hugging Face Hub (rwlinno/GSA_volc). Metadata imported from the dataset’s Hub tags.