Grounding DINO receives a photo and a phrase about the subject. The model returns match scores and frames that can be verified on the original frame.
Category that can be called
A shelf photo shows a red mug, a can, and a glass resembling the mug. Grounding DINO takes a category phrase and returns candidate bounding boxes. Start with a short phrase like the ones in the official examples and overlay the boxes on the original photo. Check matches manually. A high score alone does not tell you which feature the model relied on.
Set up the threshold
Photograph several versions of the shelf. In one the object is clearly visible, in another it is partly blocked, and in a third it is absent. Add a similarly colored object with a different shape. Run the model with the same phrase. Change the threshold and compare missed objects and extra boxes. You can remove extra boxes by hand when preparing annotations. Automated counts may need a different balance.
Keep the result for a frame with several mugs separate. One good match does not tell you whether the others were found.
From result to markup
If the frame is needed for subsequent processing, write down the size of the input image and the method of recalculation of coordinates. To accurately separate the object from the background, you will need contour marking or a segmentation model. When changing a phrase, keep the results of both formulations and repeat the control on an empty frame. This helps to note that the broader word has increased the number of matches due to similar items.
Suitable tasks
- Find in the photo items called words
- Compare gaps and extra frames at different thresholds
Strengths
- Searches for the object by the passed phrase
- Developer publishes code, weights and examples
Limitations
- A bounding box locates an object without tracing its outline
- Wording and thresholds can change the number and quality of matches
Selection and conditions
- Free access
- Yes
- Trial period
- Not confirmed
- No sign-up required
- Not confirmed
- Russian interface
- Not confirmed
- Local installation
- Yes
- API
- Not confirmed
“Not confirmed” means there is not enough information. The feature may still be available; check the developer’s terms.
- Input formats
- Image, Text
- Result
- Data
Terms
See the repository and model card for the terms covering the weights and code. Local and cloud runs both require compute. A bounding box locates an object without tracing its outline.From task to first result
Practical example
- Task
- Find in the photo items called words
- What to prepare
- Training staff shelves with two types of items and frame without the desired view.
Task brief or prompt
Find red mug objects. Save the coordinates, estimates and specified thresholds for each frame.
Expected result
Verifiable limits and a list of omissions and false matches.
How to check the result
- Compare similar items separately.
- Check the empty frame and change the threshold.
An editorial teaching example of a task brief, not the result of hands-on testing of the service. Check that the required mode is available, that you have rights to the source material and that the intended use of the result is permitted.
Check details
Official sources
This description is based on the developer’s material. Features, prices and availability may change; check the primary source before use.