ProPicker

Installation and Setup

1. Use Conda to install the necessary dependencies in a new environment. From the root of the repository, run:

   conda env create -f environment.yml
   conda activate ppicker

2. Install ProPicker itself:

   pip install .

3. You need the checkpoint of our pre-trained model, as well as the checkpoint of the TomoTwin model we used as prompt encoder.

   • You can download the ProPicker checkpoint here here

   • You can download the TomoTwin checkpoint by running bash download_tomotwin_ckpt.sh

   After downloading, place the files in the ProPicker directory.

4. Set the following environment variables to point to the model files:

   export PROPICKER_MODEL_FILE=/abs/path/to/propicker.ckpt
   export TOMOTWIN_MODEL_FILE=/abs/path/to/tomotwin.pth

Usage

There are two main workflows: prompt-based picking and fine-tuning. We provide tutorials for both.

Workflow 1: Prompt-Based Picking

Prompt-based picking can be done via both the Python package and the GUI/CLI. There are separate tutorials for both:

• Python Package Tutorial: See the TUTORIAL1:empiar10988_prompt_based_picking.ipynb notebook.

• GUI/CLI Tutorial: See the our gui_tutorial.

Workflow 2: Fine-Tuning ProPicker

Fine-tuning currently only works via the Python package and additional scripts. See the TUTORIAL2:empiar10988_fine_tuning.ipynb notebook, which continues from the first tutorial.

Training ProPicker from Scratch

Training from scratch lives in propicker/training_from_scratch/train.py, with parameters in propicker/training_from_scratch/train_cfg.py.

To download the training data, you can use datasets/download_train_data.sh.

Note: The training data is large, so you might want to download it to a different location. To do this, modify datasets/download_train_data.sh; also adjust the training data path in propicker/paths.py.

Note

This repository contains code copied and modified from the following projects:

• DeepETPicker
• DeePiCt
• TomoTwin

All derived code is explicitly marked as such.