DeepCell-Spots (Custom Fork for Cluster Use)

This is a customized fork of deepcell-spots for running DeepCell spot detection in parallel on compute clusters (e.g., SLURM arrays).

---

Custom Patch: Modified spot_detection.py

This fork includes a modification to deepcell_spots/applications/spot_detection.py to prevent issues when extracting the model archive concurrently in job arrays.

What Changed?

• The model is no longer auto-extracted by every job.
• Prevents file corruption like PermissionError and Read less bytes than requested.
• Compatible with SLURM parallel processing of FOV/Z combinations.

---

Bypass DeepCell Access Token

Instead of requiring a DEEPCELL_ACCESS_TOKEN, pre-extract the model manually:

mkdir -p ~/.deepcell/models/SpotDetection-8
tar -xzf ~/.deepcell/models/SpotDetection-8.tar.gz -C ~/.deepcell/models/SpotDetection-8

---

Running deepcell_spots_experiment

1. Clone This Fork

git clone https://github.com/tsuijenk/DeepCell-Spots.git
cd DeepCell-Spots

Ensure your environment uses this version of the package.

2. SLURM Execution

Run DeepCell Spots Per-FOV/Z (Non-Preprocessed Images):

sbatch ./experimental_run_script/polaris_run_both_xps.sh

Merge Output per Dataset:

sbatch ./experimental_run_script/batch_merge_polaris_output.sh

---

Note on Model Tarball and Permissions

You do not need to change file permissions on SpotDetection-8.tar.gz.

In this patched version of deepcell-spots, the model is not auto-extracted if the extracted folder already exists. This avoids concurrent extraction issues.

To set it up correctly:

mkdir -p ~/.deepcell/models/SpotDetection-8
tar -xzf ~/.deepcell/models/SpotDetection-8.tar.gz -C ~/.deepcell/models/SpotDetection-8

Make sure the following files and directories exist inside ~/.deepcell/models/SpotDetection-8/:

• saved_model.pb
• keras_metadata.pb
• variables/
• assets/

No access token is required once the model is extracted.