DinoSam
Bases: ModelBase
__init__(sam_model_id='facebook/sam2.1-hiera-large', dino_model_id='IDEA-Research/grounding-dino-base', box_threshold=0.3, text_threshold=0.3, *args, **kwargs)
A model combining SAM2 and GroundingDINO for promptable instance segmentation. Inspired by LangSAM and SamGeo.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
sam_model_id
|
str
|
SAM2 model. Possible options include 'facebook/sam2.1-hiera-tiny', 'facebook/sam2.1-hiera-small' and 'facebook/sam2.1-hiera-large'. Defaults to 'sam2.1-hiera-large'. |
'facebook/sam2.1-hiera-large'
|
dino_model_id
|
str
|
A GroundingDINO model. Defaults to "IDEA-Research/grounding-dino-base" |
'IDEA-Research/grounding-dino-base'
|
box_threshold
|
float
|
This parameter is used for modulating the identification of objects in the images. The box threshold is related to the model confidence, so a higher value makes the model more selective because it is equivalent to requiring the model to only select objects that it feels confident about. Defaults to 0.3. |
0.3
|
text_threshold
|
float
|
This parameter is also used for influencing the selectivity of the model by requiring a stronger association between the prompt and the segment. Defaults to 0.3. |
0.3
|