Prompt

messages = [
    {
        "role": "user",
        "content": [
            {"type": "image_url", "image_url": convert_to_base64(element_image)},
            {
                "type": "text",
                "text": """Analyze the provided cropped image from a screenshot to determine whether it contains a single, valid, and visually complete UI element.

                Criteria for validity:
                - The image must contain exactly one UI element.
                - The element must be entirely visible within the cropped area, with no significant cut-off parts.
                - The image should not consist solely of background, empty space, or meaningless fragments.

                Response format:
                Conclude with your final determination in a dedicated section:

                Conclusion
                Yes (if the image contains a single, valid, and complete UI element)
                No (if it does not meet the criteria)"""
            }
        ]
    }
]

Result Parsing

last_lines = response.strip().split('\n')[-2:]
conclusion = " ".join(last_lines).lower()
is_valid_box = "yes" in conclusion
Downloads last month
230
Safetensors
Model size
8.29B params
Tensor type
BF16
ยท
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for likaixin/UI-Box-Scorer-Qwen2.5-VL-7B

Quantizations
2 models