Add paper and code links, and update task category

README.md

@@ -1,21 +1,24 @@
 ---
 language:
-  - en
+- en
+size_categories:
+- n<1K
 task_categories:
-  - visual-question-answering
+- video-text-to-text
 pretty_name: FastBench
-size_categories:
-  - n<1K
 tags:
-  - video
-  - streaming-video
-  - video-question-answering
-  - temporal-reasoning
-  - benchmark
+- video
+- streaming-video
+- video-question-answering
+- temporal-reasoning
+- benchmark
 ---
 
 # FastBench: Can Streaming VLMs Perceive High-Dynamic Real-World Streams?
 
+**Paper:** [FastBench: Can Streaming VLMs Perceive High-Dynamic Real-World Streams?](https://huggingface.co/papers/2610.12427)  
+**Code:** [https://github.com/Ashone3/FastBench](https://github.com/Ashone3/FastBench)
+
 FastBench evaluates high-dynamic perception in streaming vision-language models. It contains **300 video clips and 306 English question-answer pairs**, with six clips containing two questions. Models must observe video incrementally, capture brief events, and answer at the appropriate time.
 
 This dataset accompanies the FastBench paper and its **ProactiveFrame** training-free adaptive frame-rate baseline. The release is an evaluation benchmark; no training or validation split is provided.
@@ -105,4 +108,4 @@ FastBench is intended for research evaluation of streaming video perception. Its
 
 The MIT license for the evaluation code does not grant rights to these videos. No dataset-wide license is declared in this card; annotation and media usage remains subject to applicable rights and source terms.
 
-When using FastBench, cite **FastBench: Can Streaming VLMs Perceive High-Dynamic Real-World Streams?** A formal citation can be added when the paper's public bibliographic details are available.
+When using FastBench, cite **FastBench: Can Streaming VLMs Perceive High-Dynamic Real-World Streams?** A formal citation can be added when the paper's public bibliographic details are available.