Update README.md

README.md

@@ -1,3 +1,84 @@
 ---
 license: apache-2.0
+language:
+- en
+library_name: transformers
+base_model:
+- Qwen/Qwen2.5-VL-7B-Instruct
+datasets:
+- YouTube-UGC
+tags:
+- video-quality-assessment
+- no-reference-vqa
+- visual-quality-r1
+- qwen2.5-vl
+- mds-vqa
+- arxiv:2603.11525
 ---
+
+# VQR1-7B-YouTubeUGC
+
+`VQR1-7B-YouTubeUGC` is the base video quality assessment (VQA) model used in **MDS-VQA: Model-Informed Data Selection for Video Quality Assessment**. It instantiates the base quality model `f(·)` in the MDS-VQA pipeline and is trained on the YouTube-UGC source dataset to predict perceptual video quality scores.
+
+MDS-VQA augments this base model with a separate failure predictor `g(·)` and a diversity-aware selection module. This repository contains the base VQA model checkpoint, not the MDS-VQA failure predictor or an actively fine-tuned target-domain model.
+
+Paper: [arXiv:2603.11525](https://arxiv.org/abs/2603.11525)  
+Project/code: [Multimedia-Analytics-Laboratory/MDS-VQA](https://github.com/Multimedia-Analytics-Laboratory/MDS-VQA)
+
+## Model Details
+
+- **Model type:** no-reference video quality assessment vision-language model
+- **Backbone family:** Qwen2.5-VL / VisualQuality-R1-style VLM
+- **Architecture in this repository:** `Qwen2_5_VLForConditionalGeneration`
+- **Parameters:** approximately 8.29B BF16 parameters
+- **Training data:** YouTube-UGC, used as the labeled source-domain dataset in MDS-VQA
+- **Input:** a video plus a VQA prompt
+- **Output:** a quality score on a 1 to 5 scale, typically inside `<answer>...</answer>` tags
+- **License:** Apache 2.0
+
+## Intended Use
+
+This model is intended for research on no-reference video quality assessment and data selection for VQA. Typical uses include:
+
+- predicting a baseline quality score for user-generated or streaming videos;
+- serving as the base quality model `f(·)` in the MDS-VQA pipeline;
+- generating baseline predictions for failure-prediction training and model-informed data selection;
+- comparing active data selection or fine-tuning methods for VQA.
+
+It is not intended as a universal production QoE monitor without domain-specific validation. Quality scores can shift across datasets, display conditions, content types, and distortion families.
+
+## Prompt Format
+
+The model follows the VisualQuality-R1-style scoring prompt used in MDS-VQA:
+
+```text
+You are doing the video quality assessment task.
+Here is the question: What is your overall rating on the quality of this video? The rating should be a float between 1 and 5, rounded to two decimal places, with 1 representing very poor quality and 5 representing excellent quality.
+First output the thinking process in <think> </think> tags and then output the final answer with only one score in <answer> </answer> tags.
+```
+
+For automatic evaluation, parse the scalar value inside the final `<answer>` tag.
+
+## Example Usage
+
+Please refer to the `src/inference.py` with the former prompt format.
+
+## MDS-VQA Context
+
+MDS-VQA is a model-informed data selection mechanism for VQA. Given an unlabeled video pool, it selects videos that are both:
+
+1. **Difficult for the base VQA model:** estimated by a failure predictor trained to rank videos by the base model's prediction errors.
+2. **Diverse in content:** estimated from frame-level semantic video features, using a Chamfer-distance-based diversity term.
+
+## Citation
+
+If you use this model, please cite MDS-VQA:
+
+```bibtex
+@article{zou2026mds,
+  title={MDS-VQA: Model-Informed Data Selection for Video Quality Assessment},
+  author={Zou, Jian and Xu, Xiaoyu and Wang, Zhihua and Wang, Yilin and Adsumilli, Balu and Ma, Kede},
+  journal={arXiv preprint arXiv:2603.11525},
+  year={2026}
+}
+```