Update model card metadata: add pipeline tag and remove arxiv tag

This PR updates the model card YAML metadata to follow Hugging Face best practices:

- Adds `pipeline_tag: image-to-image` so the model is discoverable under the correct task category.
- Removes the `arxiv:2610.17639` tag from the metadata, since arxiv IDs should be referenced in the Markdown content rather than in YAML.

The paper is already linked in the Markdown content, so no additional paper link is needed. No other content changes are made.

README.md

@@ -1,15 +1,15 @@
 ---
-license: apache-2.0
 base_model: Qwen/Qwen3.5-0.8B
-base_model_relation: finetune
 language:
-  - en
+- en
+license: apache-2.0
+pipeline_tag: image-to-image
 tags:
-  - image-cropping
-  - aesthetic-image-cropping
-  - computer-vision
-  - vision-language-model
-  - arxiv:2610.17639
+- image-cropping
+- aesthetic-image-cropping
+- computer-vision
+- vision-language-model
+base_model_relation: finetune
 ---
 
 > **New official model repository:** [zlab-sjtu/FrameLite](https://huggingface.co/zlab-sjtu/FrameLite). Please use this repository for model downloads and updates. This location remains available for existing links.
@@ -159,4 +159,4 @@ FrameLite task weights and inference code are provided under the [Apache License
 
 We thank the developers of [Qwen](https://github.com/QwenLM), [Transformers](https://github.com/huggingface/transformers), and [PEFT](https://github.com/huggingface/peft).
 
-Correspondence: [Yulun Zhang](mailto:yulun100@gmail.com).
+Correspondence: [Yulun Zhang](mailto:yulun100@gmail.com).