FastSAM

Overviews

The Fast Segment Anything Model(FastSAM) is a CNN Segment Anything Model trained using only 2% of the SA-1B dataset published by SAM authors.

Example FastSAM

Original git: https://github.com/CASIA-IVA-Lab/FastSAM.git

In this document, we will show you how to use HACHIX version of these algorithms.

Install Requirements

Please install ecos_core-1.3.0 Install from zip:

pip install ecos_core-1.3.0.zip

Configs

opt.json like this:

{
    "weights_url": "https://drive.google.com/u/0/uc?id=1m1sjY4ihXBU1fZXdQ-Xdj-mDltW-2Rqv&export=download&confirm=t&uuid=ab4e4a93-cfc4-4118-ab4f-0e1ea2d8be79&at=ALt4Tm1VVjRzBfbguE6B1gL8aVf4:1689323952061",
    "retina": true,
    "imgsz": 640,
    "conf": 0.25,
    "iou": 0.7,
    "box_prompt": "[[0,0,0,0]]",
    "text_prompt": "",
    "point_prompt": "[[0,0]]",
    "point_label": "[0]",
    "withContours": false,
    "better_quality": false,
    "text_conf": 0.25,
    "clip_model_name": "ViT-B/32"

}

• weights_url: Weight url for download
• retina: Draw high-resolution segmentation masks. Value: true or false
• imgsz: image size.
• conf: confidence threshold.
• iou: IoU threshold.
• box_prompt: Support multiple boxes [[x,y,w,h],[x2,y2,w2,h2]]. Default: "[[0,0,0,0]]"
• text_prompt: use text prompt eg: "a dog". Default: ""
• point_prompt: Point to detect [[x1,y1],[x2,y2]]. Default: "[[0,0]]"
• point_label: 0:background, 1:foreground. Default: "[0]"
• withContours: Draw the edges of the masks. Value: true or false
• better_quality: Better quality using morphologyEx. Value: true or false
• text_conf: text confidence
• clip_model_name: model name of CLIP

Priority output

Boxes (box_prompt) -> Text (text_promt)-> Point (point_prompt and point_label)-> All masks

If you want to select type of detections, please set other parameters as default value.

APIs

Model

Bases: BaseModel

This is the base model for INDAD

__init__(opt=None)

This function is called when creating a new instance of FastSAM Model

Parameters:

Name	Type	Description	Default
opt	NameSpace	The default options neccessary for initializing the model. opt should includes these attributes: weight.	None

build_model(custom_weight=None)

build model for detection

Returns:

Name	Type	Description
Model		Inherited from model base class

get_transform()

Get image data transform function for preprocessing

Returns:

Name	Type	Description
transform	func	transform function. This function will take input as image path and output - raw_image: image numpy mat - image_transform: image tensor (pytorch)

predict(input)

Predict

Parameters:

Name	Type	Description	Default
input	(URI, OpenCV, PIL, numpy)	input	required

Returns:

Name	Type	Description
Detections		Output

process_predictions(pred, raw_image_path, raw_image_mat, image_transform, save_image_path)

post process the output of the net

Parameters:

Name	Type	Description	Default
pred	_type_	output of detection	required
raw_image_path	str	raw image path	required