title	author	date
Change Detection SAR to Data Cube	Denis Berroeta	2023-10-30

Convencional Methods

Segmenting remote sensing imagery with FastSAM

[📕Paper] [🤗HuggingFace Demo] [Colab demo] [Replicate demo & API] [Model Zoo] [BibTeX]

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. FastSAM achieves comparable performance with the SAM method at 50× higher run-time speed. Reference: https://samgeo.gishub.org/examples/fast_sam/

Installation

Clone the repository locally:

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

Create the conda env. The code requires python>=3.7, as well as pytorch>=1.7 and torchvision>=0.8. Please follow the instructions here to install both PyTorch and TorchVision dependencies. Installing both PyTorch and TorchVision with CUDA support is strongly recommended.

conda create -n FastSAM python=3.9
conda activate FastSAM

Install the packages:

cd FastSAM
pip install -r requirements.txt

Install CLIP:

pip install git+https://github.com/openai/CLIP.git

Model Checkpoints

Two model versions of the model are available with different sizes. Click the links below to download the checkpoint for the corresponding model type.

• default or FastSAM: YOLOv8x based Segment Anything Model | Baidu Cloud (pwd: 0000).
• FastSAM-s: YOLOv8s based Segment Anything Model.

Usage

Step 1: Image preprocessing

preprocessing unique:

To transform image to be read by FastSAM then apply the utils/transform_image.py script, for example:

apply_transform(input_image_file='./images/ST_001/DD_ST_001_255.tif',
                output_image_file='./sam_imputs/ST_001/DD_ST_001_255.tif')

Multi preprocessing:

If you want to apply the above procedure massively to all image files contained in folders and subfolders you should use the utils/transform_fsam_sp.py script and add only the source folder (input_folder) and the output folder (output_folder).

...

# Iterar por multiples carpetas
input_folder = "MC_DIF_samsara"
output_folder = 'sam_input'

...

Step 2: Apply FastSAM

This procedure returns the segmentation in different formats:

• Image PNG
• Image TIF
• Polygon Shapefile
• Polygon geojson

unique segmentation:

The second step is the segmentation with the FastSAM model, where you can apply the model for an image using the fastSAM_sp.py script, just modify the path of the input image, as shown below:

...

# model fastSAM
model_path = 'weights/FastSAM-x.pt'
sam = SamGeo(model=model_path)

image = "sam_imputs/ST_007/DD_ST_007.tif"
sam.set_image(image)
...

Multi segmentation:

If you want to apply the above procedure in bulk to all image files contained in folders and subfolders you should use the fastSAM_all.py script and add only the source folder (input_folder) and the output folder (output_folder).

...

# Modelos SAM
model_path = 'weights/FastSAM-x.pt'
sam = SamGeo(model=model_path)

# Iterar por multiples carpetas
input_folder = "sam_input"
output_folder = 'results'
...