Skip to content

Installation Guide

Requirements

Saber runs on Python 3.10 and above on Linux or Windows with CUDA12.

Quick Installation

Saber is available on PyPI and can be installed using pip:

pip install saber-em

⚠️ Note By default, the GUI is not included in the base installation. To enable the graphical interface for manual annotation, install with: 

pip install saber-em[gui]

Development Installation

If you want to contribute to saber or need the latest development version, you can install from source:

git clone https://github.com/chanzuckerberg/saber.git
cd saber
pip install -e .

Verification

To verify your installation, run:

saber

You should see the following output:

 SABER ⚔️ -- Segment Anything Based Expert             
 Recognition.                                                
╭─ Options ──────────────────────────────────────────╮
│ --help  -h  Show this message and exit.            │
╰────────────────────────────────────────────────────╯
╭─ Commands ─────────────────────────────────────────╮
│ analysis    Post-processing analysis after         │
│             segmentation.                          │
│ classifier  Routines for training and evaluating   │
│             classifiers.                           │
│ gui         Saber GUI for annotating SAM2          │
│             segmentations with custom classes.     │
│ save        Save organelle coordinates and         │
│             statistics (size-distribution) from    │
│             segmentations.                         │
│ segment     Segment Tomograms and Micrographs with │
│             SABER.                                 │
│ web         SABER Annotation GUI Web Server        │
╰────────────────────────────────────────────────────╯

Troubleshooting

HuggingFace Authentication Error (GatedRepoError)

If you see the following error when running Saber:

GatedRepoError: 401 Client Error.
Cannot access gated repo for url https://huggingface.co/facebook/sam3/resolve/main/config.json.
Access to model facebook/sam3 is restricted. You must have access to it and be authenticated to access it. Please log in.

This means the SAM3 model weights are hosted in a gated HuggingFace repository that requires you to explicitly request access. To fix this:

  1. Create or log in to your HuggingFace account.
  2. Visit the facebook/sam3 model page and request access.
  3. Once access is granted, authenticate your local environment: 
    pip install huggingface_hub
hf auth login
    When prompted, enter your HuggingFace access token (available at huggingface.co/settings/tokens).
  4. Re-run your Saber command — the model weights will now download automatically.

Next Steps