[Feature]: solution to items that requires very small threshold in order for grounding dino to detect: manually draw box prompt in addition to box from grounding dino/show which box is selected when box selection by user is changed

[Inertia05]

Expected behavior

I noticed that something specfic is very hard for groundingdino to detect. for example the engine exhaust from a jet engine in a big picture with 3 planes flying overhead in background. if the threshold is high, the grounding dino cannot detect it. if threshold is low, the grounding dino generate at least 20 boxes that all have number overlapped and it's very hard to identify the correct box number. example below:

[continue-revolution]

I’ve already had plan to resolve this issue, but since it’s very front-end (tons of javascript, which I’m not familiar with), it will be implemented after some backend improvements