StevenHH2000/iNat21-1shot-fewshots

--- dataset_info: features: - name: image sequence: image - name: problem dtype: string - name: label dtype: string - name: solution sequence: string - name: image_width dtype: int64 - name: image_height sequence: int64 task_categories: - image-text-to-text language: - en library_name: datasets tags: - multimodal - llm - vision --- This is the official release of the training dataset for paper **[Taxonomy-Aware Representation Alignment for Hierarchical Visual Recognition with Large Multimodal Models](https://arxiv.org/abs/2603.00431)**. Code is available at [https://github.com/PKU-ICST-MIPL/TARA_CVPR2026](https://github.com/PKU-ICST-MIPL/TARA_CVPR2026). # Data Source ## **Training** - We randomly sample 1-shot data per category from iNaturalist2021 dataset. # Data Fields - **image:** input image(s) - data type: dict - **problem:** input question - data type: string - **label:** coarse-to-fine categories - data type: string - **image_width:** image width - data type: int64 - **image_height:** image height - data type: int64