napari-sam3-assistant 4.2.3

A napari plugin for Segment Anything Model 3 (SAM3) image segmentation with Simple and Advanced workflows, text, points, boxes, exemplars, large-image ROI inference, mask operations, and 3D/video-like propagation

napari-sam3-assistant

napari-sam3-assistant is a napari plugin for Segment Anything Model 3 (SAM3) image segmentation. Version 4 adds a two-mode interface: Simple for guided image segmentation and Advanced for the original full Step 1 to Step 6 workflow.

The plugin focuses on task-based segmentation workflows:

• 2D segmentation with text, box, point, and mask-style prompts
• 3D stack / video-like propagation from prompts on a selected slice or frame
• exemplar segmentation from Shapes ROI boxes
• text-based concept segmentation
• large OME-Zarr and TIFF segmentation through local ROI inference
• Live Points with positive and negative prompts
• downstream mask cleanup, merge, and export operations

What's New in 4.2.3 SAM3 video diagnostics and performance logging

Version 4.2.3 adds an opt-in diagnostic mode for SAM3.1 video and 3D propagation performance troubleshooting.

• Added a Log SAM3.1 diagnostics checkbox in Advanced > Step 2. Task Setup > Advanced.
• When enabled, the plugin records detailed SAM3 video timing and runtime diagnostics during model loading, session startup, prompt insertion, propagation, and napari label writing.
• The diagnostic mode is intended for comparing SAM3.1 multiplex performance across machines, CUDA/PyTorch builds, and plugin execution paths.
• The option is off by default so normal users are not exposed to extra debug logging during routine segmentation.

What's New in 4.2.2 Optional completion chime for long runs

SAM3 Assistant can play a short, soft completion chime when a long-running task finishes. This is useful when preview or 3D/video propagation takes more than one minute and the user is working away from the screen. The chime is optional and can be turned on or off from the plugin UI. On Windows, SAM3 Assistant uses the native winsound backend first for more reliable playback, while Linux and macOS keep the standard plugin sound path with a fallback beep if needed.

What's New in 4.2.0

Version 4.2.0 adds experimental CPU-only support for SAM3.0 2D image workflows when using a CPU-safe sam3 backend such as sam3-cpu.

• CPU-only setup is documented in docs/cpu_only.md.
• Device mode is now environment-driven and shown as an indicator; normal users no longer need to choose CPU or GPU manually.
• Advanced manual device override is available only for backend testing with NAPARI_SAM3_ENABLE_DEVICE_OVERRIDE=1.
• SAM3.0 2D CPU workflows are enabled for points, boxes, text, exemplar, and Live Points when the installed sam3 backend supports CPU model construction.
• SAM3.1, 3D/video propagation, and SAM3 video-predictor workflows remain CUDA/GPU-only in this plugin.
• Model-folder setup now passes the detected BPE tokenizer path to SAM3 and can create bpe_simple_vocab_16e6.txt.gz automatically from merges.txt.
• The plugin reports a clear upstream CPU-support limitation if a non-CPU-safe SAM3 backend still allocates CUDA tensors during CPU image model construction.

What's New in 4.0.0

Version 4.0.0 was a workflow release focused on the new Simple mode and a cleaner Advanced mode.

• New Simple mode for common imaging tasks with a compact one-column layout.
• Advanced mode keeps the full manual UI for model setup, batch work, large-image ROI settings, result tables, mask operations, and detailed logs.
• The mode selector stays visible, so users can move between Simple and Advanced without restarting napari.
• Simple mode uses the same SAM3 execution path and writes the same napari preview layers as Advanced mode.
• Simple mode keeps common tasks short: choose the image/task, add or enter the prompt, then run preview.
• Simple mode includes Mask Ops in the Run area to open the standalone mask cleanup widget for preview labels.
• Simple mode uses SAM3.0 for 2D image tasks so Advanced SAM3.1 video-model settings do not break Simple image segmentation.
• Device selection is explicit. GPU / CUDA is recommended for full SAM3 functionality; CPU is experimental for SAM3.0 2D image workflows and requires a CPU-safe SAM3 backend.
• Live Points are still available with T for next point mode and Shift+T to flip selected or latest points.

SAM 3 is not bundled with this plugin. Install the SAM 3 backend and download the SAM 3 model files separately from Meta's Hugging Face repository.

Status

This project is under active development. The current widget supports local SAM 3 model loading, napari prompt collection, Simple and Advanced UI modes, large-image ROI execution, downstream mask operations, background execution, and writing results back to napari layers.

Changelog

Release notes and bug-fix history are maintained in CHANGELOG.md.

Requirements

• Python >=3.11
• napari >=0.5
• SAM 3 Python package importable as sam3
• CUDA-enabled PyTorch and torchvision installed for your platform for normal use
• A local SAM 3 checkpoint directory containing:
  • config.json
  • processor_config.json
  • one weight file such as sam3.pt, model.safetensors, or sam3.1_multiplex.pt

CPU-only use is possible for SAM3.0 2D image workflows with a CPU-safe SAM3 backend. The tested optional backend is the external rhubarb-ai/sam3-cpu fork, which reports version 0.1.0 and is distributed under the MIT license in its own repository. It is not bundled with this plugin. See CPU-only SAM3.0 setup.

Setup

Windows

1. Download and install Miniforge:
   https://github.com/conda-forge/miniforge/releases/latest/download/Miniforge3-Windows-x86_64.exe

2. After Miniforge is installed, open either:

   • Miniforge Prompt
   • PowerShell

3. Create and activate the environment

conda create -n napari-sam3 python=3.11 -y
conda activate napari-sam3

4. Install base Python tools and napari

python -m pip install --upgrade pip wheel
python -m pip install "setuptools<82"
python -m pip install "napari[all]"

5. Install CUDA-enabled PyTorch

python -m pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu128

6. Install SAM3

Choose one:

Option A. Install from a local clone

git clone https://github.com/facebookresearch/sam3.git
cd sam3
python -m pip install --no-cache-dir -e .

Option B. Install directly from GitHub

python -m pip install --no-cache-dir git+https://github.com/facebookresearch/sam3

7. Install extra dependencies

python -m pip install einops triton-windows pycocotools

If SAM3.1 multiplex propagation later fails on Windows with errors such as No available kernel. Aborting execution!, see the replacement-file workaround in windows_sam31_workaround/README.md.

8. Install napari-sam3-assistant

python -m pip install napari-sam3-assistant

If you are installing from a local repository checkout instead:

python -m pip install -e .

9. Launch napari

napari

Linux ARM64 (AArch64)

1. Download Miniforge https://conda-forge.org/download/

Install Miniforge first.

chmod +x Miniforge3-Linux-aarch64.sh
./Miniforge3-Linux-aarch64.sh -b -p "$HOME/miniforge3"
source "$HOME/miniforge3/bin/activate"

2. Create and activate the environment

conda create -n napari-sam3 python=3.11 -y
conda activate napari-sam3

3. Install base Python tools and napari

python -m pip install --upgrade pip wheel
python -m pip install "setuptools<82" "numpy>=1.26,<2"
python -m pip install "napari[bermuda, pyqt6, optional-numba, optional-base]"

4. Install CUDA-enabled PyTorch

python -m pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu130

5. Install SAM3

git clone https://github.com/facebookresearch/sam3.git
cd sam3
python -m pip install --no-cache-dir -e .

6. Install extra dependencies

python -m pip install einops triton pycocotools

7. Install napari-sam3-assistant

python -m pip install napari-sam3-assistant

If you are installing from a local repository checkout instead:

python -m pip install -e .

8. Launch napari

napari

Verify the installation

Inside the activated environment, confirm that SAM3 imports correctly:

python -c "import sam3; print('sam3 import OK')"

You can also verify napari:

python -c "import napari; print(napari.__version__)"

Download SAM 3 model files

Download the model files from:

• https://huggingface.co/facebook/sam3
• https://huggingface.co/facebook/sam3.1

These repositories are gated, so you must request or accept access first.

After approval, download the files manually from the Files and versions tab.

A reference screenshot of the file list is shown below:

Model folder

The model files can be stored in any folder you want.

Examples:

D:\models\sam3
D:\models\sam3_1

The plugin only needs the correct folder path.

Example SAM3.0 folder:

D:\models\sam3\
  config.json
  processor_config.json
  sam3.pt

model.safetensors is also supported as a SAM3 3.0 weight file.

Example SAM3.1 folder:

D:\models\sam3_1\
  config.json
  processor_config.json
  sam3.1_multiplex.pt

Current model support:

• SAM3.0 weights: 2D image tasks and 3D/video propagation
• SAM3.1 sam3.1_multiplex.pt: 3D/video propagation through the SAM3.1 multiplex video predictor
• SAM3.1 is not currently routed through the plugin's 2D image model loader

Device rule

• Use a CUDA-enabled PyTorch environment for normal use.
• Select GPU / CUDA for SAM3.0, SAM3.1, and 3D/video workflows.
• CPU is experimental for SAM3.0 2D image workflows with a CPU-safe SAM3 backend. Tested CPU workflows include points, boxes, exemplar, text, and Live Points. See CPU-only SAM3.0 setup.

For developers

If you want editable installs and git pull, use:

git clone https://github.com/facebookresearch/sam3.git
cd sam3
python -m pip install --no-cache-dir -e .

git clone https://github.com/wulinteousa2-hash/napari-sam3-assistant.git
cd napari-sam3-assistant
python -m pip install -e .

Basic Workflow

Simple Mode

Use Simple when you want a guided image-segmentation workflow with fewer controls on screen.

The main flow is:

1. Select an image and choose the task.
2. Add the prompt.
3. Click Run Preview.

Simple mode is intended for common imaging tasks:

• 2D: points, boxes, labels-mask, or text prompts on the selected image plane. CPU mode requires a CPU-safe SAM3 backend.
• Text: enter a short imaging concept such as cell, nucleus, or myelin
• Refine: use Live Points to add positive or negative point corrections
• Exemplar: draw one or more boxes around example objects
• 3D/Video: start propagation from a prompt on the selected frame or slice when the current data and model support it

Simple mode keeps model setup small:

• model folder
• GPU / CUDA or experimental CPU
• SAM3.0 for Simple image tasks

Use Advanced when you need SAM3.1 model selection, batch processing, large-image ROI controls, detailed result tables, or CSV export.

After a Simple preview creates labels, click Mask Ops in the Run area to open SAM3 Mask Operations on the cleanup tab.

Advanced Mode

Advanced is the original full workflow. It keeps the Step 1 to Step 6 layout for users who want manual control.

1. Open an image in napari.
2. Open Plugins > SAM3 Assistant.
3. Choose Advanced.
4. Select the image in Target image.
5. Select a task.
6. Create a prompt layer if the task needs one.
7. Click Run Preview.
8. Inspect SAM3 preview labels, SAM3 preview masks, or SAM3 preview boxes.
9. Use Save & Clean in Step 4. Run and Save for quick mask acquisition, or open SAM3 Mask Operations if you want advanced cleanup, merge, or export controls.

Use Clear Preview to remove generated preview layers without deleting prompts or saved labels.

Run and Save

Step 4. Run and Save includes a quick save path for users who want to acquire a mask and immediately continue segmentation.

After a preview is created, choose:

• output folder
• output format
• filename

Then click Save & Clean.

Save & Clean:

• saves the current preview mask as a new napari Labels layer
• writes the mask file to the selected output folder
• removes temporary preview layers
• releases Python / CUDA temporary memory where available
• unloads the SAM3 model so memory returns closer to baseline
• leaves an Open Folder shortcut for the saved mask location

If Load model when running is checked, the next run reloads the model automatically.

Supported quick-save formats:

Preview type	TIFF	NumPy .npy	PNG
2D mask	Yes	Yes	Yes
3D/video propagated mask	Yes	Yes	No

PNG is only offered for 2D masks. For 3D/video outputs, use TIFF stack or NumPy .npy.

Large-Image Local Inference

Large-image mode is optional and off by default. When it is off, the plugin keeps the existing full-image inference path.

Use this mode for OME-Zarr, large TIFF, and similar large images where sending the full selected plane to SAM3 is too expensive.

Workflow:

1. Set up the normal task and prompt type.
2. Enable Enable large-image local inference in Advanced task setup.
3. Choose a local ROI size:

512 x 512
1024 x 1024
2048 x 2048
4096 x 4096
8192 x 8192

4. Add a point or box prompt.
5. Click Run Preview, or add points in Live Points mode.

ROI behavior:

• Point prompts use the latest point as the ROI anchor.
• Box prompts use the box center and keep the box inside the local inference window when possible.
• Live Points use the latest point as the ROI anchor.
• Text-only prompts keep the full-image path in this first pass unless a point or box anchor is also available.
• If a new point or box stays inside the current ROI, the same ROI is reused.
• If a new point or box falls outside the current ROI, the ROI is rebuilt around the new prompt.

The active ROI is shown as:

SAM3 active ROI

SAM3 receives only the local ROI image data. Returned labels and boxes are written back into global image coordinates in the normal preview layers.

Status messages report:

Large-image mode OFF: full-image inference.
Large-image mode ON: local ROI inference (WIDTH x HEIGHT).
Active ROI bounds: y=Y0:Y1, x=X0:X1.

For large ROI work, Save & Clean is the recommended handoff after a successful preview. It saves the mask, clears temporary preview memory, unloads the model, and lets users continue with another ROI without manually visiting Mask Operations.

Batch Multiple 2D Images

Use Batch all image layers when several open 2D images should receive the same prompt setup.

Workflow:

1. Open multiple images in napari.
2. Configure a 2D task such as text, box, exemplar, or labels-mask segmentation.
3. Add the prompt once.
4. Enable Batch all image layers in Advanced.
5. Click Run Preview.

Each source image gets its own output layers:

SAM3 preview labels [image name]
SAM3 preview masks [image name]
SAM3 preview boxes [image name]

Batch preview layers can be reviewed directly or used in Mask Operations.

Batch mode is intended for 2D image tasks. It is disabled for Live Points and 3D/video propagation because those workflows depend on one active image/session.

Multi-Text Batch Mode

Use Batch text prompts when you want each text concept to run independently instead of writing one combined phrase such as cat and dog.

Workflow:

1. Set Task to Text segmentation.
2. Enter one concept per line in Batch text prompts:

cat
dog
person

3. Leave Batch all image layers off to run all prompts on the selected image only.
4. Enable Batch all image layers to run every prompt on every open image.
5. Click Run Preview.

Outputs include both image and prompt:

SAM3 preview labels [Image 1 - cat]
SAM3 preview labels [Image 1 - dog]
SAM3 preview labels [Image 2 - cat]
SAM3 preview labels [Image 2 - dog]

The Results table includes a Prompt column. Object IDs are scoped to each image-prompt result, so Object ID 1 for Image 1 - cat is separate from Object ID 1 for Image 1 - dog.

Tasks

Text Segmentation

Use text to segment all matching instances of a concept.

Workflow:

1. Set Task to Text segmentation.
2. Leave Prompt type as Text only.
3. Enter a short phrase, for example:

cell
nucleus
myelin
myelin sheath

4. Keep Detection threshold near the default 0.35, or lower it if the result is empty.
5. Press Enter in the text prompt field or click Run Preview.

No prompt layer is needed for text segmentation. Create Prompt Layer is not required.

Text prompts usually work better as short noun phrases than instructions. Prefer myelin sheath over segment all the myelin rings. The plugin strips common instruction prefixes before sending the prompt to SAM3, but microscopy-specific language can still be difficult for the model.

If the result says objects=0, SAM3 ran but did not return masks above threshold. Try a shorter noun phrase, lower Detection threshold, or use a box/exemplar prompt for structures that are visually clear but not well recognized by text.

2D Segmentation With Boxes

Use boxes to segment the target region inside each drawn box.

Workflow:

1. Set Task to 2D segmentation.
2. Set Prompt type to Box.
3. Click Create Prompt Layer.
4. Draw one or more rectangles in the SAM3 boxes Shapes layer.
5. Click Run Preview.

Each 2D box preview writes segmentation only inside the corresponding drawn box. This is different from Exemplar segmentation, which uses boxed examples to find and segment similar objects outside the original boxes.

Exemplar Segmentation

Use example ROIs to segment similar objects.

Workflow:

1. Set Task to Exemplar segmentation.
2. Set Prompt type to Box.
3. Click Create Prompt Layer.
4. Draw boxes around one or more example objects.
5. Click Run Preview.

The local SAM 3 image API exposes visual exemplars through geometric box prompts. The plugin stores ROI metadata, but inference currently passes exemplar ROIs as SAM 3 visual box prompts.

Exemplar segmentation is a 2D/image task in this plugin. For exemplar-like 3D propagation, use 3D/video propagation with a box prompt on the selected frame or slice.

Live Points With Positive and Negative Points

Use points to correct a result.

Workflow:

1. Set Task to Live Points.
2. Set Prompt type to Points (positive/negative).
3. Click Create Prompt Layer.
4. Choose Positive and add points on regions to include.
5. Choose Negative and add points on regions to exclude.
6. Click Run Preview.

This is useful after a text, box, or exemplar preview is close but not correct.

Labels Mask Prompt

Use a napari Labels layer as a mask-style prompt.

Workflow:

1. Set a task that supports mask prompts.
2. Set Prompt type to Labels mask.
3. Click Create Prompt Layer.
4. Paint non-zero pixels in SAM3 mask prompt.
5. Click Run Preview.

Labels-mask prompts are currently supported for 2D/image workflows, not 3D/video propagation. For 3D/video propagation, use text, box, or point prompts on the selected frame.

3D Stack / Video Propagation

Treat a stack as video-like data and propagate a prompt through frames or slices.

Workflow:

1. Open a stack in napari.
2. Set Task to 3D/video propagation.
3. Select the target frame or slice in napari.
4. Create a prompt layer and add prompts on that frame.
5. Choose propagation direction:
   • both
   • forward
   • backward
6. Click Start 3D Propagation.

In 3D/video propagation mode, the primary run button changes from Run Preview to Start 3D Propagation. This starts a new SAM3 video session, adds the current frame prompt, and propagates through the stack. Propagate Existing Session is an advanced action that reuses the current SAM3 video session without adding a new prompt; it is enabled only after a successful 3D propagation run.

3D/video prompt limits:

• SAM3.0 video propagation supports one initial visual box on the prompted frame.
• SAM3.1 video multiplex can accept multiple box prompts.
• Point prompts target one object per request and cannot be mixed with text or box prompts in the same 3D/video request.
• Point prompts are limited to 16 points per request. SAM3's tracker prompt encoder uses the first 8 and last 8 points when more are supplied, so the plugin fails early instead of letting middle points be ignored.
• Labels-mask prompts are not supported by the SAM3 video predictor API used by this plugin.
• Exemplar segmentation itself is routed through the 2D image model. Use box prompts in 3D/video propagation when the goal is exemplar-like propagation through a stack.

Preview output is written to:

SAM3 propagated preview labels

Saved output is written to:

SAM3 saved propagated labels

The current SAM 3 video predictor backend is CUDA-only. CPU mode is experimental for SAM3.0 2D image workflows with a CPU-safe SAM3 backend; see CPU-only SAM3.0 setup.

Channel Axis

Channel axis tells the plugin which data axis is color/channel.

Default:

-1

Use -1 for grayscale images and normal RGB/RGBA images. The plugin auto-detects trailing RGB/RGBA axes of size 3 or 4.

Examples:

(H, W)          -> -1
(H, W, 3)      -> -1
(H, W, 4)      -> -1
(Z, H, W)      -> -1
(C, H, W)      -> 0
(Z, C, H, W)   -> 1
(T, C, H, W)   -> 1
(Z, H, W, C)   -> 3

Leave it at -1 unless your image has an explicit multi-channel microscopy dimension.

Output Layers

Preview layers:

SAM3 preview labels
SAM3 preview masks
SAM3 preview boxes
SAM3 propagated preview labels

Saved layers:

SAM3 saved labels
SAM3 saved propagated labels

Buttons:

• Validate: check the selected SAM 3 model directory.
• Load 2D Model: load the 2D/image model.
• Load 3D/Video Model: load the video propagation model.
• Run Preview: run the selected task.
• Clear Preview: remove generated preview layers only.
• Cancel: stop a running worker.
• Unload: unload the SAM3 model from memory.
• Save Accepted Object: save a preview label object in Mask Operations.

Results table:

Layer | Prompt | Frame | Object ID | Score | Area

• Layer: source image layer.
• Prompt: text prompt used for text and multi-text results. For non-text workflows this is -.
• Frame: propagated frame or slice index. For 2D results this is -.
• Object ID: SAM3 object ID when available, otherwise a generated label ID.
• Score: SAM3 confidence/probability when returned by the backend.
• Area: number of mask pixels for that object in the displayed 2D plane or frame.

Results actions:

• Clear Results: clear the table only.
• Copy Clipboard: copy tab-separated results, including headers, for pasting into Excel or statistics software.
• Export CSV: save the results table to a CSV file.

Tested Image Coverage

The following image sizes and channel layouts have been exercised with the current workflow:

Image type	Size / layout	Workflow	Status
Single-channel 2D	about 2048 x 2048	2D preview and quick save	Passed
RGB 2D	about 2048 x 2048 x 3	2D preview and quick save	Passed
Single-channel large image	about 60000 x 60000	large-image local ROI inference with quick save and clean	Passed
RGB large image	about 60000 x 60000 x 3	large-image local ROI inference	Not yet tested

Large-image coverage means the plugin was used with local ROI inference rather than sending the full image plane to SAM3 at once. Actual memory use depends on ROI size, model type, device, source image backend, and whether napari already holds large arrays in memory.

Label-value merge:

Use this when multiple SAM3 objects should become the same class value in a Labels layer. For example, if labels 3, 4, 5, and 6 are all the same biological class, set:

Values to replace: 3,4,5,6
New value: 3

Then click Merge Label Values. The selected Labels layer is updated in place.

Mask Operations

SAM3 Mask Operations is a separate napari widget for turning SAM3 previews into curated masks for analysis or training data. Open it from the plugin menu or from the Mask Ops button in Simple mode. When opened from Mask Ops, it starts as a floating tool window.

Tabs:

• Accepted Objects: save a preview Labels layer as a named accepted object with class metadata, append it to an existing accepted layer, or replace an existing accepted layer.
• Class Merge: merge selected accepted-object layers into a class working mask.
• Mask Cleanup: analyze connected components, delete selected components, remove small objects, fill holes, smooth masks, keep the largest component, and relabel values.
• Final Merge / Export: merge cleaned class masks into semantic, instance, or binary final masks, choose overlap handling, and export TIFF, PNG, or NumPy .npy files.

The mask operations panel works on napari Labels layers, including SAM3 preview and saved label layers.

Mouse-assisted cleanup:

• In Mask Cleanup, enable Right-click Delete.
• Right-click a label object in the selected target Labels layer to remove that label value.
• Click Undo Last Edit to restore the previous mask state for the selected Labels layer.
• Double-click a component table row to jump the viewer to that mask component.
• This is useful for supervised cleanup after SAM3 creates a preview mask.

Overlap inspection:

• In Final Merge / Export, select two or more class mask layers.
• Click Show Overlap Map.
• The plugin creates SAM3 overlap map, where non-zero pixels mark locations covered by more than one selected class mask.

ARM64, CUDA, and DGX Spark

For ARM64 systems such as NVIDIA DGX Spark / GB10:

• Use Python 3.11 or newer.
• Keep the NVIDIA driver and CUDA stack current.
• Install a PyTorch/torchvision build that supports your GPU architecture.
• Use a PyTorch/torchvision/SAM3 build with CUDA kernels compatible with the device.
• CPU-only PyTorch requires a CPU-safe SAM3 backend for image workflows; see CPU-only SAM3.0 setup.

Check PyTorch GPU support:

python - <<'PY'
import torch
print("torch:", torch.__version__)
print("torch cuda runtime:", torch.version.cuda)
print("cuda available:", torch.cuda.is_available())
if torch.cuda.is_available():
    print("device:", torch.cuda.get_device_name(0))
    print("capability:", torch.cuda.get_device_capability(0))
    print("arch list:", torch.cuda.get_arch_list())
PY

GB10 reports compute capability 12.1 (sm_121). If your PyTorch build does not include compatible kernels, you may see:

CUDA error: no kernel image is available for execution on the device
nvrtc: error: invalid value for --gpu-architecture

The plugin does not compile PyTorch, torchvision, or SAM 3 CUDA extensions.

Troubleshooting

No mask appears and status says objects=0

SAM 3 returned no detections above threshold. Try:

• a shorter text prompt
• a more common concept phrase
• a lower Detection threshold
• box or exemplar prompts
• a CUDA/PyTorch/SAM3 build compatible with your GPU

CUDA kernel image error

Error:

CUDA error: no kernel image is available for execution on the device

The GPU is visible, but at least one required CUDA kernel was not built for the device architecture. Install compatible PyTorch/torchvision/SAM 3 builds for the GPU. For CPU-only 2D use, see CPU-only SAM3.0 setup.

Invalid GPU architecture

Error:

nvrtc: error: invalid value for --gpu-architecture

The installed PyTorch CUDA runtime cannot compile for the detected GPU. Install a PyTorch/torchvision/SAM 3 build that supports the GPU. For CPU-only 2D use, see CPU-only SAM3.0 setup.

BFloat16 conversion errors

The plugin converts SAM3 bfloat16 outputs to float32 before writing NumPy-backed napari layers. If you still see dtype errors, restart napari after changing device mode and run again.

SAM3.1 start_session fails with unexpected keyword argument 'offload_state_to_cpu'

This is a plugin/backend API mismatch, not a prompt or data problem.

• The failure happens during start_session / init_state, before propagation actually begins.
• The installed sam3 backend does not accept the keyword that newer plugin code may pass.
• The plugin includes compatibility handling for this case so older installed sam3 backends can still start a 3D/video session.

If you still see this exact error, first verify that napari is importing the intended local sam3 install and not an older duplicate environment copy.

SAM3.1 propagation fails later with No available kernel. Aborting execution! on Windows

This is a different issue from the offload_state_to_cpu startup mismatch.

• start_session succeeds.
• prompts are accepted.
• the failure happens only when SAM3.1 multiplex propagation actually starts.

On some Windows systems using triton-windows, this appears to be an upstream sam3 runtime/kernel path issue rather than a napari prompt-collection bug.

Use the documented Windows workaround here:

• windows_sam31_workaround/README.md

Text prompt creates no layer

That is expected. Text segmentation does not need a prompt layer. Enter text and click Run Preview.

Development

Install in editable mode:

pip install -e .

Run tests:

PYTHONPATH=src pytest -q

The test suite covers coordinate mapping, prompt collection, adapter utility behavior, and static widget UI checks. It does not download SAM 3 weights.

References

• SAM 3 repository: https://github.com/facebookresearch/sam3
• SAM 3 model files: https://huggingface.co/facebook/sam3
• PyTorch installation selector: https://pytorch.org/get-started/locally/

Acknowledgement

The demo image was provided by the Electron Microscopy Core Facility at Houston Methodist Research Institute

License

MIT. See the project license file.