There are many failure examples when the prompt is long

[HuiZhang0812]

Hello, thank you for your excellent work. However, when I use a detail prompt as the input for grounding dino 1.5, there is consistently no bounding box in the output. When I switch to a shorter prompt, the output includes appropriate bounding boxes. Is this phenomenon expected? And is there any way to resolve this issue?

Original image:

Output using a detailed prompt, which lacks any bounding boxes:

The prompt is: "This is a photo showcasing a classical-style building and its staircase. The staircase is located in the center of the image, composed of stone steps and a railing, with a stone base at the bottom. The walls of the building are covered with green ivy, adding a touch of nature. The windows of the building are neatly arranged, with a row of white curtains hanging above. The entire scene is illuminated by natural light, presenting a tranquil and ancient atmosphere."

Output using a concise prompt, which includes appropriate bounding boxes:

The prompt is: "building.staircase.green ivy.windows." The shorter prompt was adapted by me to follow the style of the demo and the paper.