Datasets:
Modalities:
Text
Formats:
parquet
Languages:
English
Size:
1K - 10K
Tags:
multimodal
vision-language-model
visual-question-answering
fine-grained-perception
on-policy-distillation
resolution
License:
| RP-OPSD Training Dataset | |
| Copyright 2026 RP-OPSD contributors | |
| This product includes selected and transformed material from the following | |
| public Hugging Face dataset releases, each identified by its direct upstream | |
| repository as licensed under the Apache License, Version 2.0: | |
| 1. LMMs-Lab-Turtle/Vision-SR1-47K | |
| Revision: 2900b038f4aaa72f6b92795c1ee3ab29b7d509b6 | |
| 2. UCSC-VLAA/VLM-CapCurriculum-Perception-Data | |
| Revision: 9f26e8d4e1b1c9d5bb53989cee4755c07f8ebb84 | |
| 3. inclusionAI/ZwZ-RL-VQA | |
| Revision: fec404cb7bd5a18458e5add2802a5fb183981d14 | |
| 4. yuanqianhao/Vision-OPD-6K | |
| Revision: eb5c1c2e7b9a7b6a619efe4161c7369c71bf8af4 | |
| Modifications in this release include ordered sample selection, deterministic | |
| prompt normalization, deterministic option formatting, removal of red-box | |
| instructions, use of original no-box images where applicable, duplicate | |
| removal, conversion to a unified Parquet schema, and creation of student | |
| images at half the original physical width and height using Lanczos resampling. | |
| The upstream project names, author names, and trademarks are used only for | |
| attribution and identification. They do not imply endorsement of RP-OPSD or | |
| this dataset release. | |