Open X-Embodiment
1M+ real-robot trajectories from 20+ institutions.
UHR-BAT-SFT-10K UHR-BAT-SFT-10K is the supervised fine-tuning dataset used for UHR-BAT: Budget-Aware Token Compression Vision-Language Model for Ultra-High-Resolution Remote Sensing. UHR-BAT targets efficient and accurate understanding of ultra-high-resolution remote-sensing imagery by allocating visual token budgets according to the current instruction and preserving query-relevant regional evidence. The dataset contains visual question answering style instruction-following samples… See the full description on the dataset page:
Source: Hugging Face Hub (RL-MIND/UHR-BAT-SFT-10K). Metadata imported from the dataset’s Hub tags.