This project focuses on determining the diameter distribution of NPLs with optimized magnetic properties colloidally stabilized in a solvent or soft matrix, utilizing the Segment Anything Model. The original README file of the SAM project can be found under README_SAM.md.
git clone git@github.com:manojlovska/segment-anything.gitconda create -n env_name python=3.8.5
conda activate env_namecd segment-anything
pip install git+https://github.com/facebookresearch/segment-anything.gitInstall PyTorch 1.7.1 (or later) and torchvision, as well as small additional dependencies. On a CUDA GPU machine, the following will do the trick:
pip install torch==2.2.1 torchvision==0.17.1 torchaudio==2.2.1 --index-url https://download.pytorch.org/whl/cu118Or if you don't have GPU 9230 s:
pip install torch==2.2.1 torchvision==0.17.1 torchaudio==2.2.1 --index-url https://download.pytorch.org/whl/cpuHere you can find the pytorch distribution that suits your hardware requirements.
pip install -r requirements.txtmkdir weights
cd weights
wget https://dl.fbaipublicfiles.com/segment_anything/sam_vit_b_01ec64.pth
cd ..Run the SAM-zero-shot.py script to obtain the masks. An example of the following command:
python SAM-zero-shot.py --img_path "/path/to/Porazdelitev delcev original" --model-type "vit_b" --device "cuda" --convert-to-rle --img-extension "tif" --checkpoint weights/sam_vit_b_01ec64.pth - Notes:
- change --img_path "/path/to/Porazdelitev delcev original" to the path to your dataset, or an image file you want segmented
- --convert-to-rle flag must be present to save the masks in coco RLE format
- the masks will be saved in "./output" by default, add --output "path/to/desired/output/directory" if you want to change this
- change --device 'cuda' to --device 'cpu' if you do not want to run the program on a GPU
Run
python SAM-zero-shot.py --helpfor additional information on how to use the command.
If you don't have ground truth diameters distributions, you can plot only the predicted diameter distributions. For example, to generate diameters distribution out of one image, run the following command:
python generate_histograms_main.py --images-path '/path/to/Porazdelitev delcev original/BSHF-DBSA-210325/BSHF-DBSA-210325_0001.tif' --masks-path "./output/BSHF-DBSA-210325/BSHF-DBSA-210325_0001.json" --save-hist --ratio 0.2410Or if you want to plot the diameter distribution of an experimental analysis consisted of multiple images:
python generate_histograms_main.py --images-path '/path/to/Porazdelitev delcev original/BSHF-DBSA-210325' --masks-path "./output/BSHF-DBSA-210325" --save-hist --ratio 0.2410Generated histograms and predicted diameters in nm are saved in "./results" directory, you can change this by specifying different --save-path.
Here you must specify the masks path and the ratio for converting the diameters in nm.
Run
python compare_histograms_main.py --images-path '/path/to/Porazdelitev delcev original' --save-histto compare the predicted and ground truth histograms for the whole dataset or
python compare_histograms_main.py --images-path '/path/to/Porazdelitev delcev original/BSHF-DBSA-210325' --save-histfor a particular experimental analysis.
- Notes:
- Run
for additional information on how to use this command.python compare_histograms_main.py --help
- If you wish you can choose different size constraints --t_min and --t_max, default are 10 nm and 150 nm
- Add --filter-edge to filter out the edge particles (Extremely slow on non-GPU machine!)
- If you chose different path to save the output masks of the SAM model, you have to specify that path here by using --masks-path
Generated histograms and predicted diameters in nm are saved in "./results" directory, you can change this by specifying different --save-path.
If you want to check if the obtained masks for an image are well generated, run the command:
python visualize_masks.py --image-path '/path/to/Porazdelitev delcev original/BSHF-DBSA-210325/BSHF-DBSA-210325_0001.tif' --save-plot --ratio 0.2410Change the size constraints by specifying --tmin and --tmax. Specify different --save-path to save the plot in different directory.
For all the scripts you can run the following command for additional information:
python script-name.py --help