### Install Supervoxel Module (Subdirectory)

Source: https://github.com/mic-dkfz/nninteractive/blob/master/AGENTS.md

Navigate to the supervoxel subdirectory and perform an editable installation.

```bash
cd nnInteractive/supervoxel/src/sam2 && pip install -e .
```

--------------------------------

### Install nnInteractive Repository

Source: https://github.com/mic-dkfz/nninteractive/blob/master/readme.md

Installs the nnInteractive repository either via pip or by cloning and installing from source.

```bash
pip install nninteractive
```

```bash
git clone https://github.com/MIC-DKFZ/nnInteractive
cd nnInteractive
pip install -e .
```

--------------------------------

### Install SAM 2 with Notebook Support

Source: https://github.com/mic-dkfz/nninteractive/blob/master/nnInteractive/supervoxel/src/sam2/INSTALL.md

Use this command to install SAM 2 and its dependencies for notebook usage. Ensure Python and PyTorch are installed beforehand.

```bash
pip install -e ".[notebooks]"
```

--------------------------------

### Install Supervoxel Module (Top-level)

Source: https://github.com/mic-dkfz/nninteractive/blob/master/AGENTS.md

Navigate to the supervoxel directory and perform an editable installation.

```bash
cd nnInteractive/supervoxel && pip install -e .
```

--------------------------------

### Install SAM 2

Source: https://github.com/mic-dkfz/nninteractive/blob/master/nnInteractive/supervoxel/src/sam2/README.md

Clone the SAM 2 repository and install it using pip. Ensure you have Python 3.10+ and PyTorch 2.5.1+ installed.

```bash
git clone https://github.com/facebookresearch/sam2.git && cd sam2

pip install -e .
```

--------------------------------

### Editable Install with Dev Tools

Source: https://github.com/mic-dkfz/nninteractive/blob/master/AGENTS.md

Install the core package along with formatter and linter tooling for development.

```bash
pip install -e ".[dev]"
```

--------------------------------

### Editable Install Core Package

Source: https://github.com/mic-dkfz/nninteractive/blob/master/AGENTS.md

Use this command for editable installation of the core package during development.

```bash
pip install -e .
```

--------------------------------

### Install nnInteractive SuperVoxel Dependencies

Source: https://github.com/mic-dkfz/nninteractive/blob/master/nnInteractive/supervoxel/README.md

Install the modified SAM2 and the SuperVoxel repository. Ensure you are in the correct directories before running pip install.

```bash
cd src/sam2/
pip install -e .
cd ../..
pip install -e .
```

--------------------------------

### Install SAM 2 Without Build Isolation

Source: https://github.com/mic-dkfz/nninteractive/blob/master/nnInteractive/supervoxel/src/sam2/INSTALL.md

Use the `--no-build-isolation` flag with pip if installation fails even after verifying CUDA setup. This can sometimes resolve issues related to build environments.

```bash
pip install --no-build-isolation -e .
```

--------------------------------

### Verify PyTorch CUDA Setup

Source: https://github.com/mic-dkfz/nninteractive/blob/master/nnInteractive/supervoxel/src/sam2/INSTALL.md

Confirm that PyTorch can detect your CUDA installation and that the CUDA_HOME variable is correctly set. This check is essential before proceeding with CUDA-dependent installations.

```python
import torch; from torch.utils.cpp_extension import CUDA_HOME; print(torch.cuda.is_available(), CUDA_HOME)
```

--------------------------------

### Launch Multi-Node Training with SLURM

Source: https://github.com/mic-dkfz/nninteractive/blob/master/nnInteractive/supervoxel/src/sam2/training/README.md

Initiate multi-node training on a cluster using SLURM. This example configures training across 2 nodes, each with 8 GPUs, and includes optional SLURM parameters.

```python
python training/train.py \
    -c configs/sam2.1_training/sam2.1_hiera_b+_MOSE_finetune.yaml \
    --use-cluster 1 \
    --num-gpus 8 \
    --num-nodes 2
    --partition $PARTITION \
    --qos $QOS \
    --account $ACCOUNT
```

--------------------------------

### Install PyTorch with CUDA Support

Source: https://github.com/mic-dkfz/nninteractive/blob/master/readme.md

Installs PyTorch with CUDA support for Ubuntu systems with Nvidia GPUs. Adjust CUDA version based on your drivers.

```bash
pip3 install torch torchvision --index-url https://download.pytorch.org/whl/cu126
```

--------------------------------

### Load SAM 2 Video Predictor from Hugging Face

Source: https://github.com/mic-dkfz/nninteractive/blob/master/nnInteractive/supervoxel/src/sam2/README.md

Loads the SAM 2 video predictor directly from Hugging Face. Requires the `huggingface_hub` library to be installed.

```python
import torch
from sam2.sam2_video_predictor import SAM2VideoPredictor

predictor = SAM2VideoPredictor.from_pretrained("facebook/sam2-hiera-large")

with torch.inference_mode(), torch.autocast("cuda", dtype=torch.bfloat16):
    state = predictor.init_state(<your_video>)

    # add new prompts and instantly get the output on the same frame
    frame_idx, object_ids, masks = predictor.add_new_points_or_box(state, <your_prompts>):

    # propagate the prompts to get masklets throughout the video
    for frame_idx, object_ids, masks in predictor.propagate_in_video(state):
        ...
```

--------------------------------

### Load SAM 2 Image Predictor from Hugging Face

Source: https://github.com/mic-dkfz/nninteractive/blob/master/nnInteractive/supervoxel/src/sam2/README.md

Loads the SAM 2 image predictor directly from Hugging Face. Requires the `huggingface_hub` library to be installed.

```python
import torch
from sam2.sam2_image_predictor import SAM2ImagePredictor

predictor = SAM2ImagePredictor.from_pretrained("facebook/sam2-hiera-large")

with torch.inference_mode(), torch.autocast("cuda", dtype=torch.bfloat16):
    predictor.set_image(<your_image>)
    masks, _, _ = predictor.predict(<input_prompts>)
```

--------------------------------

### Create Conda Virtual Environment

Source: https://github.com/mic-dkfz/nninteractive/blob/master/readme.md

Creates and activates a new Conda virtual environment for nnInteractive. Ensure Python 3.10+ is installed.

```bash
conda create -n nnInteractive python=3.12
conda activate nnInteractive
```

--------------------------------

### Force Build SAM 2 CUDA Extension

Source: https://github.com/mic-dkfz/nninteractive/blob/master/nnInteractive/supervoxel/src/sam2/INSTALL.md

Reinstall SAM 2 and force the build of the CUDA extension by setting SAM2_BUILD_ALLOW_ERRORS to 0. This will raise errors if the build fails. Ensure PyTorch and matching CUDA toolkits are installed first.

```bash
pip uninstall -y SAM-2 && \
rm -f ./sam2/*.so && \
SAM2_BUILD_ALLOW_ERRORS=0 pip install -v -e ".[notebooks]"
```

--------------------------------

### Install SAM 2 Skipping CUDA Extension

Source: https://github.com/mic-dkfz/nninteractive/blob/master/nnInteractive/supervoxel/src/sam2/INSTALL.md

Install SAM 2 without building the CUDA extension by setting the SAM2_BUILD_CUDA environment variable to 0. This skips the post-processing step at runtime.

```bash
SAM2_BUILD_CUDA=0 pip install -e ".[notebooks]"
```

--------------------------------

### Verify SAM 2 Base Module Path

Source: https://github.com/mic-dkfz/nninteractive/blob/master/nnInteractive/supervoxel/src/sam2/INSTALL.md

Check if the `sam2_base.py` file is up-to-date by printing its path and comparing it with the latest version in the repository. This helps identify if an outdated installation is causing errors.

```python
from sam2.modeling import sam2_base

print(sam2_base.__file__)
```

--------------------------------

### SAM 2 Image Prediction

Source: https://github.com/mic-dkfz/nninteractive/blob/master/nnInteractive/supervoxel/src/sam2/README.md

Initialize the SAM2ImagePredictor with a specified model configuration and checkpoint, then set an image and perform prediction with input prompts. This example requires PyTorch and runs in inference mode with CUDA and bfloat16 precision.

```python
import torch
from sam2.build_sam import build_sam2
from sam2.sam2_image_predictor import SAM2ImagePredictor

checkpoint = "./checkpoints/sam2.1_hiera_large.pt"
model_cfg = "configs/sam2.1/sam2.1_hiera_l.yaml"
predictor = SAM2ImagePredictor(build_sam2(model_cfg, checkpoint))

with torch.inference_mode(), torch.autocast("cuda", dtype=torch.bfloat16):
    predictor.set_image(<your_image>)
    masks, _, _ = predictor.predict(<input_prompts>)
```

--------------------------------

### Specify CUDA Capability for Compilation

Source: https://github.com/mic-dkfz/nninteractive/blob/master/nnInteractive/supervoxel/src/sam2/INSTALL.md

Use this command to manually specify the CUDA capability when compiling to match your GPU, especially if the installation environment differs from the runtime environment.

```bash
export TORCH_CUDA_ARCH_LIST=9.0 8.0 8.6 8.9 7.0 7.2 7.5 6.0
```

--------------------------------

### Set CUDA Home Environment Variable

Source: https://github.com/mic-dkfz/nninteractive/blob/master/nnInteractive/supervoxel/src/sam2/INSTALL.md

Explicitly set the CUDA_HOME environment variable to the path of your CUDA toolkit installation if the build process cannot find it. This is crucial for compiling CUDA kernels.

```bash
export CUDA_HOME=/usr/local/cuda  # change to your CUDA toolkit path
```

--------------------------------

### Initialize nnInteractive Inference Session

Source: https://github.com/mic-dkfz/nninteractive/blob/master/readme.md

Sets up an nnInteractive inference session, downloads model weights, and loads the trained model. Ensure the download directory exists and has sufficient space.

```python
import os
import torch
import SimpleITK as sitk
from huggingface_hub import snapshot_download  # Install huggingface_hub if not already installed

# --- Download Trained Model Weights (~400MB) ---
# License reminder: The official nnInteractive checkpoint is licensed under
# Creative Commons Attribution Non Commercial Share Alike 4.0 (CC BY-NC-SA 4.0).
# See the License section of this readme!.
REPO_ID = "nnInteractive/nnInteractive"
MODEL_NAME = "nnInteractive_v1.0"  # Updated models may be available in the future
DOWNLOAD_DIR = "/home/isensee/temp"  # Specify the download directory

download_path = snapshot_download(
    repo_id=REPO_ID,
    allow_patterns=[f"{MODEL_NAME}/*"],
    local_dir=DOWNLOAD_DIR
)

# The model is now stored in DOWNLOAD_DIR/MODEL_NAME.

# --- Initialize Inference Session ---
from nnInteractive.inference.inference_session import nnInteractiveInferenceSession

session = nnInteractiveInferenceSession(
    device=torch.device("cuda:0"),  # Set inference device
    use_torch_compile=False,  # Experimental: Not tested yet
    verbose=False,
    torch_n_threads=os.cpu_count(),  # Use available CPU cores
    do_autozoom=True,  # Enables AutoZoom for better patching
    use_pinned_memory=True,  # Optimizes GPU memory transfers
)

# Load the trained model
model_path = os.path.join(DOWNLOAD_DIR, MODEL_NAME)
session.initialize_from_trained_model_folder(model_path)

# --- Load Input Image (Example with SimpleITK) ---
# DO NOT preprocess the image in any way. Give it to nnInteractive as it is! DO NOT apply level window, DO NOT normalize 
# intensities and never ever convert an image with higher precision (float32, uint16, etc) to uint8!
# The ONLY instance where some preprocesing makes sense is if your original image is too large to be reasonably used. 
# This may be the case, for example, for some microCT images. In this case you can consider downsampling.
input_image = sitk.ReadImage("FILENAME")
img = sitk.GetArrayFromImage(input_image)[None]  # Ensure shape (1, x, y, z)

# Validate input dimensions
if img.ndim != 4:
    raise ValueError("Input image must be 4D with shape (1, x, y, z)")

session.set_image(img)
```

--------------------------------

### Generate nnUNet-Compatible Foreground Prompts

Source: https://github.com/mic-dkfz/nninteractive/blob/master/nnInteractive/supervoxel/README.md

Create .pkl files containing foreground locations from generated supervoxels for nnUNet training. Specify the supervoxel mask folder and the number of parallel processes.

```bash
SuperVoxel_save_fg_location \
  -supervoxel_folder /path/to/supervoxel/masks \
  -np 4  # number of parallel processes
```

--------------------------------

### Initialize and Use Video Predictor

Source: https://github.com/mic-dkfz/nninteractive/blob/master/nnInteractive/supervoxel/src/sam2/README.md

This snippet shows how to initialize the SAM 2 video predictor, add prompts, and propagate masks throughout a video.

```APIDOC
## Initialize and Use Video Predictor

### Description
This example demonstrates initializing the SAM 2 video predictor, setting up the inference state, adding new prompts (points or boxes) to a frame, and then propagating these prompts to generate masks throughout the video.

### Usage
```python
import torch
from sam2.build_sam import build_sam2_video_predictor

checkpoint = "./checkpoints/sam2.1_hiera_large.pt"
model_cfg = "configs/sam2.1/sam2.1_hiera_l.yaml"
predictor = build_sam2_video_predictor(model_cfg, checkpoint)

with torch.inference_mode(), torch.autocast("cuda", dtype=torch.bfloat16):
    # Initialize the state with your video data
    state = predictor.init_state(<your_video>)

    # Add new prompts (points or boxes) to the current frame and get immediate output
    frame_idx, object_ids, masks = predictor.add_new_points_or_box(state, <your_prompts>)

    # Propagate the prompts to generate masklets throughout the video
    for frame_idx, object_ids, masks in predictor.propagate_in_video(state):
        # Process the masks for each frame
        ...
```

### Parameters
- `state`: The inference state object, initialized by `init_state`.
- `<your_video>`: Placeholder for the video data to be processed.
- `<your_prompts>`: Placeholder for the user-provided prompts (e.g., points, boxes).
```

--------------------------------

### Run All Pre-commit Hooks

Source: https://github.com/mic-dkfz/nninteractive/blob/master/AGENTS.md

Execute all configured pre-commit hooks locally to ensure code quality and consistency before committing.

```bash
pre-commit run --all-files
```

--------------------------------

### Launch Single-Node Training

Source: https://github.com/mic-dkfz/nninteractive/blob/master/nnInteractive/supervoxel/src/sam2/training/README.md

Execute the training script for single-node fine-tuning on the MOSE dataset using 8 GPUs. The '--use-cluster 0' flag indicates non-cluster execution.

```python
python training/train.py \
    -c configs/sam2.1_training/sam2.1_hiera_b+_MOSE_finetune.yaml \
    --use-cluster 0 \
    --num-gpus 8
```

--------------------------------

### Initialize SAM 2 Video Predictor from Checkpoint

Source: https://github.com/mic-dkfz/nninteractive/blob/master/nnInteractive/supervoxel/src/sam2/README.md

Initializes the SAM 2 video predictor using a local checkpoint and configuration file. Ensure you have the necessary checkpoint and model configuration files.

```python
import torch
from sam2.build_sam import build_sam2_video_predictor

checkpoint = "./checkpoints/sam2.1_hiera_large.pt"
model_cfg = "configs/sam2.1/sam2.1_hiera_l.yaml"
predictor = build_sam2_video_predictor(model_cfg, checkpoint)
```

--------------------------------

### Set New Image and Target Buffer

Source: https://github.com/mic-dkfz/nninteractive/blob/master/readme.md

Load a new image and set a corresponding target buffer. The new target buffer must match the shape of the new image.

```python
session.set_image(NEW_IMAGE)
session.set_target_buffer(torch.zeros(NEW_IMAGE.shape[1:], dtype=torch.uint8))
```

--------------------------------

### Configure Mixed Image and Video Datasets for Training

Source: https://github.com/mic-dkfz/nninteractive/blob/master/nnInteractive/supervoxel/src/sam2/training/README.md

Use this configuration to train on a mix of image and video datasets. Ensure SA-V videos are extracted to JPEG frames beforehand. Adjust paths and parameters as needed.

```yaml
data:
  train:
    _target_: training.dataset.sam2_datasets.TorchTrainMixedDataset 
    phases_per_epoch: ${phases_per_epoch} # Chunks a single epoch into smaller phases
    batch_sizes: # List of batch sizes corresponding to each dataset
    - ${bs1} # Batch size of dataset 1
    - ${bs2} # Batch size of dataset 2
    datasets:
    # SA1B as an example of an image dataset
    - _target_: training.dataset.vos_dataset.VOSDataset
      training: true
      video_dataset:
        _target_: training.dataset.vos_raw_dataset.SA1BRawDataset
        img_folder: ${path_to_img_folder}
        gt_folder: ${path_to_gt_folder}
        file_list_txt: ${path_to_train_filelist} # Optional
      sampler:
        _target_: training.dataset.vos_sampler.RandomUniformSampler
        num_frames: 1
        max_num_objects: ${max_num_objects_per_image}
      transforms: ${image_transforms}
    # SA-V as an example of a video dataset
    - _target_: training.dataset.vos_dataset.VOSDataset
      training: true
      video_dataset:
        _target_: training.dataset.vos_raw_dataset.JSONRawDataset
        img_folder: ${path_to_img_folder}
        gt_folder: ${path_to_gt_folder}
        file_list_txt: ${path_to_train_filelist} # Optional
        ann_every: 4
      sampler:
        _target_: training.dataset.vos_sampler.RandomUniformSampler
        num_frames: 8 # Number of frames per video
        max_num_objects: ${max_num_objects_per_video}
        reverse_time_prob: ${reverse_time_prob} # probability to reverse video
      transforms: ${video_transforms}
    shuffle: True
    num_workers: ${num_train_workers}
    pin_memory: True
    drop_last: True
    collate_fn:
    _target_: training.utils.data_utils.collate_fn
    _partial_: true
    dict_key: all

```

--------------------------------

### Download SAM 2 Checkpoints

Source: https://github.com/mic-dkfz/nninteractive/blob/master/nnInteractive/supervoxel/src/sam2/README.md

Download all model checkpoints for SAM 2 by running the provided shell script in the checkpoints directory.

```bash
cd checkpoints && \
./download_ckpts.sh && \
cd ..
```

--------------------------------

### Configure MOSE Dataset Paths

Source: https://github.com/mic-dkfz/nninteractive/blob/master/nnInteractive/supervoxel/src/sam2/training/README.md

Set the paths for the MOSE dataset within the configuration file. Ensure 'img_folder' and 'gt_folder' point to the correct dataset locations.

```yaml
dataset:
    # PATHS to Dataset
    img_folder: null # PATH to MOSE JPEGImages folder
    gt_folder: null # PATH to MOSE Annotations folder
    file_list_txt: null # Optional PATH to filelist containing a subset of videos to be used for training
```

--------------------------------

### Rebuild SAM 2 Extensions In-Place

Source: https://github.com/mic-dkfz/nninteractive/blob/master/nnInteractive/supervoxel/src/sam2/INSTALL.md

If you encounter import errors, try rebuilding the SAM 2 extensions in the repository root. This command compiles C++ extensions locally.

```bash
python setup.py build_ext --inplace
```

--------------------------------

### Configure Experiment Log Directory

Source: https://github.com/mic-dkfz/nninteractive/blob/master/nnInteractive/supervoxel/src/sam2/training/README.md

Specify a custom path for the experiment log directory. If set to null, logs default to './sam2_logs/${config_name}'.

```yaml
experiment_log_dir: null # Path to log directory, defaults to ./sam2_logs/${config_name}
```

--------------------------------

### Add -allow-unsupported-compiler Argument

Source: https://github.com/mic-dkfz/nninteractive/blob/master/nnInteractive/supervoxel/src/sam2/INSTALL.md

Add the '-allow-unsupported-compiler' argument to `nvcc` in `setup.py` to fix compilation errors caused by incompatible CUDA and Visual Studio versions.

```python
compile_args = {
        "cxx": [],
        "nvcc": [
            "-DCUDA_HAS_FP16=1",
            "-D__CUDA_NO_HALF_OPERATORS__",
            "-D__CUDA_NO_HALF_CONVERSIONS__",
            "-D__CUDA_NO_HALF2_OPERATORS__",
            "-allow-unsupported-compiler"  # Add this argument
        ],
    }
```

--------------------------------

### Format Python Code

Source: https://github.com/mic-dkfz/nninteractive/blob/master/AGENTS.md

Format Python code in the nnInteractive directory using Black.

```bash
black nnInteractive/
```

--------------------------------

### Add SAM 2 Repository to PYTHONPATH

Source: https://github.com/mic-dkfz/nninteractive/blob/master/nnInteractive/supervoxel/src/sam2/INSTALL.md

Manually add the SAM 2 repository root to your PYTHONPATH environment variable if the `sam2` package is not found. This ensures Python can locate configuration files.

```bash
export SAM2_REPO_ROOT=/path/to/sam2  # path to this repo
export PYTHONPATH="${SAM2_REPO_ROOT}:${PYTHONPATH}"
```

--------------------------------

### Generate SuperVoxels

Source: https://github.com/mic-dkfz/nninteractive/blob/master/nnInteractive/supervoxel/README.md

Command to generate supervoxels from input images. Specify input image directory, output folder, and configuration file path. This process is compute and VRAM intensive.

```bash
SuperVoxel_generate \
  -i /path/to/input/images \
  -o /path/to/output/folder \
  -c /path/to/config.yaml
```

--------------------------------

### Video Prediction with Prompts and Propagation

Source: https://github.com/mic-dkfz/nninteractive/blob/master/nnInteractive/supervoxel/src/sam2/README.md

Performs video prediction by initializing state, adding prompts, and propagating masks throughout the video. Uses inference mode and autocast for performance.

```python
with torch.inference_mode(), torch.autocast("cuda", dtype=torch.bfloat16):
    state = predictor.init_state(<your_video>)

    # add new prompts and instantly get the output on the same frame
    frame_idx, object_ids, masks = predictor.add_new_points_or_box(state, <your_prompts>):

    # propagate the prompts to get masklets throughout the video
    for frame_idx, object_ids, masks in predictor.propagate_in_video(state):
        ...
```

--------------------------------

### Load Video Predictor from Hugging Face

Source: https://github.com/mic-dkfz/nninteractive/blob/master/nnInteractive/supervoxel/src/sam2/README.md

Load the SAM 2 video predictor directly from Hugging Face model repository.

```APIDOC
## Load Video Predictor from Hugging Face

### Description
This example shows how to load a pre-trained SAM 2 video predictor model directly from the Hugging Face Hub using the `from_pretrained` method. This simplifies the process of getting started with video prediction.

### Usage
```python
import torch
from sam2.sam2_video_predictor import SAM2VideoPredictor

# Load the predictor from Hugging Face
predictor = SAM2VideoPredictor.from_pretrained("facebook/sam2-hiera-large")

with torch.inference_mode(), torch.autocast("cuda", dtype=torch.bfloat16):
    # Initialize the state with your video data
    state = predictor.init_state(<your_video>)

    # Add new prompts and propagate masks as shown in the previous example
    frame_idx, object_ids, masks = predictor.add_new_points_or_box(state, <your_prompts>)
    for frame_idx, object_ids, masks in predictor.propagate_in_video(state):
        ...
```

### Parameters
- `"facebook/sam2-hiera-large"`: The identifier for the SAM 2 model on Hugging Face Hub.
```

--------------------------------

### Relax Attention Kernel Settings

Source: https://github.com/mic-dkfz/nninteractive/blob/master/nnInteractive/supervoxel/src/sam2/INSTALL.md

Replace the line in `transformer.py` to relax attention kernel settings and use kernels other than Flash Attention, which can resolve 'RuntimeError: No available kernel' errors.

```python
OLD_GPU, USE_FLASH_ATTN, MATH_KERNEL_ON = True, True, True
```

--------------------------------

### Add Scribble Interaction

Source: https://github.com/mic-dkfz/nninteractive/blob/master/readme.md

Incorporate a scribble interaction using a 3D image where a single slice contains a hand-drawn scribble (background 0, scribble 1). Use session.preferred_scribble_thickness for optimal results.

```python
session.add_scribble_interaction(SCRIBBLE_IMAGE, include_interaction=True)
```

--------------------------------

### Lint and Auto-fix Python Code

Source: https://github.com/mic-dkfz/nninteractive/blob/master/AGENTS.md

Run Ruff lint checks on the nnInteractive directory and automatically fix issues where possible.

```bash
ruff check nnInteractive/ --fix
```

--------------------------------

### Retrieve Segmentation Results

Source: https://github.com/mic-dkfz/nninteractive/blob/master/readme.md

Obtain the segmentation results by cloning the target buffer. Cloning is necessary as the buffer will be reused for subsequent objects.

```python
# The target buffer holds the segmentation result.
results = session.target_buffer.clone()
# OR (equivalent)
results = target_tensor.clone()

# Cloning is required because the buffer will be **reused** for the next object.
# Alternatively, set a new target buffer for each object:
session.set_target_buffer(torch.zeros(img.shape[1:], dtype=torch.uint8))
```

--------------------------------

### Define Output Buffer for Segmentation

Source: https://github.com/mic-dkfz/nninteractive/blob/master/readme.md

Initialize a zero-filled tensor with the same shape as the input image (excluding channels) and specify the data type as uint8. This tensor will serve as the target buffer for segmentation masks.

```python
target_tensor = torch.zeros(img.shape[1:], dtype=torch.uint8)  # Must be 3D (x, y, z)
session.set_target_buffer(target_tensor)
```

--------------------------------

### Check Spelling in Source/Docs

Source: https://github.com/mic-dkfz/nninteractive/blob/master/AGENTS.md

Use codespell to check for spelling errors in source code and documentation, skipping specified file types.

```bash
codespell --skip='.git,*.pdf,*.svg'
```

--------------------------------

### Add Lasso Interaction

Source: https://github.com/mic-dkfz/nninteractive/blob/master/readme.md

Incorporate a lasso interaction using a 3D image with a single slice containing a closed contour representing the selection.

```python
session.add_lasso_interaction(LASSO_IMAGE, include_interaction=True)
```

--------------------------------

### Modify PyTorch Version Restriction

Source: https://github.com/mic-dkfz/nninteractive/blob/master/nnInteractive/supervoxel/src/sam2/INSTALL.md

If facing dependency conflicts related to PyTorch versions, consider changing the version restriction in `pyproject.toml` and `setup.py` to `torch==2.1.0`. This may resolve issues caused by incompatible PyTorch versions.

```toml
torch==2.1.0
```

--------------------------------

### Reset Interactions for New Object

Source: https://github.com/mic-dkfz/nninteractive/blob/master/readme.md

Clear the target buffer and reset all interactions to begin segmenting a new object.

```python
session.reset_interactions()  # Clears the target buffer and resets interactions
```

--------------------------------

### Add Bounding Box Interaction

Source: https://github.com/mic-dkfz/nninteractive/blob/master/readme.md

Incorporate a bounding box interaction for segmentation refinement. BBOX_COORDINATES should be [[x1, x2], [y1, y2], [z1, z2]]. Note that current pre-trained models primarily support 2D bounding boxes, requiring one dimension to be [d, d+1].

```python
# Example of a 2D bounding box in the axial plane (XY slice at depth Z)
# BBOX_COORDINATES = [[30, 80], [40, 100], [10, 11]]  # X: 30-80, Y: 40-100, Z: slice 10
session.add_bbox_interaction(BBOX_COORDINATES, include_interaction=True)
```

--------------------------------

### Add Negative Point Interaction

Source: https://github.com/mic-dkfz/nninteractive/blob/master/readme.md

Add a negative point interaction by setting include_interaction=False. This helps exclude areas from the segmentation.

```python
session.add_point_interaction(POINT_COORDINATES, include_interaction=False)
```

--------------------------------

### Add Positive Point Interaction

Source: https://github.com/mic-dkfz/nninteractive/blob/master/readme.md

Incorporate a positive point interaction to refine the segmentation mask. Ensure POINT_COORDINATES is a tuple (x, y, z) representing the point's location.

```python
session.add_point_interaction(POINT_COORDINATES, include_interaction=True)
```

=== COMPLETE CONTENT === This response contains all available snippets from this library. No additional content exists. Do not make further requests.