Shape2Force / README.md
kaveh's picture
added
2b9ff22
|
raw
history blame
2.02 kB

Shape2Force (S2F)

Predict force maps from bright-field microscopy images of single-cell or spheroid using deep learning.

Web App: The app is published to Hugging Face Spaces. To work on it locally: git clone git@hf.co:spaces/kaveh/Shape2force S2FApp


Quick Start

Web app (local):

cd S2FApp
pip install -r requirements.txt
streamlit run app.py

Or use the online app on Hugging Face. Place checkpoints (.pth) in S2FApp/ckp/ for local use; the Space downloads them automatically.


Ways to Use S2F

1. Web App

Run the Streamlit GUI from S2FApp/:

cd S2FApp && streamlit run app.py
  1. Choose Model type: Single cell or Spheroid
  2. Select a Checkpoint from ckp/
  3. For single-cell: pick Substrate (e.g. fibroblasts_PDMS)
  4. Upload an image or pick from sample/
  5. Click Run prediction

Output: heatmap, cell force (sum), and basic stats.


2. Jupyter Notebook

For interactive usage and custom analysis, you may use the notebook:

  • notebooks/evaluate_model.ipynb – Load data, run evaluation, plot predictions, and save per-sample metrics.

Once cloned the repo. open the notebook in Jupyter and adjust the configuration cell (paths, model type, substrate).


3. Training & Fine-Tuning

Dataset layout: A folder with train/ and test/ subfolders. Each subfolder has:

  • BF_001.tif (bright-field image)
  • *_gray.jpg (force map / heatmap)
  • Optional .txt (cell_area, sum_force)

Single-cell:

python -m training.train --data path/to/dataset --model single_cell --epochs 100 --substrate fibroblasts_PDMS

Spheroid:

python -m training.train --data path/to/dataset --model spheroid --epochs 100

Resume / fine-tune from checkpoint:

python -m training.train --data path/to/dataset --model single_cell --resume ckp/last_checkpoint.pth --epochs 150