### Full YOLO-MLX Setup

Source: https://github.com/thewebai/yolo-mlx/blob/main/README.md

Installs YOLO-MLX and its tracking dependencies, setting up a virtual environment.

```bash
cd yolo-mlx

# Create and activate virtual environment
python3 -m venv .venv
source .venv/bin/activate

# Install the package
pip install -e .

# Install tracking dependencies (OpenCV, lap, scipy — required for model.track())
pip install -e ".[tracking]"
```

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

### Setup and Model Conversion for Training

Source: https://github.com/thewebai/yolo-mlx/blob/main/README.md

Prepares the environment and converts a YOLOv26 model to MLX format, serving as starting weights for fine-tuning.

```bash
# 1. Setup (if not done already)
cd yolo-mlx
python3 -m venv .venv && source .venv/bin/activate
pip install -e .
pip install -e ".[convert]"

# 2. Download and convert a pretrained model as starting weights
bash scripts/download_yolo26_models.sh
yolo-mlx converters convert models/yolo26n.pt -o models/yolo26n.npz --verify
```

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

### Setup and Model Conversion for Detection

Source: https://github.com/thewebai/yolo-mlx/blob/main/README.md

Installs YOLO-MLX with tracking and conversion dependencies, then downloads and converts a pretrained YOLO26 model for detection.

```bash
cd yolo-mlx
python3 -m venv .venv && source .venv/bin/activate
pip install -e .
pip install -e ".[tracking]"
pip install -e ".[convert]"
bash scripts/download_yolo26_models.sh
yolo-mlx converters convert models/yolo26n.pt -o models/yolo26n.npz --verify
```

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

### Setup and Model Conversion for Tracking

Source: https://github.com/thewebai/yolo-mlx/blob/main/README.md

Installs necessary tracking dependencies and converts a YOLOv26 model to MLX format for multi-object tracking.

```bash
# 1. Setup (if not done already)
cd yolo-mlx
python3 -m venv .venv && source .venv/bin/activate
pip install -e .
pip install -e ".[tracking]"
pip install -e ".[convert]"

# 2. Download and convert a model
bash scripts/download_yolo26_models.sh
yolo-mlx converters convert models/yolo26n.pt -o models/yolo26n.npz --verify

# 3. Download MOT17 and create a sample pedestrian video (~3s, 1080p)
bash scripts/download_mot17.sh
python scripts/create_sample_video.py       # creates images/pedestrians.mp4
```

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

### Setup and Model Conversion for Segmentation

Source: https://github.com/thewebai/yolo-mlx/blob/main/README.md

Installs YOLO-MLX with segmentation and conversion dependencies, then downloads and converts a pretrained segmentation model.

```bash
# 1. Setup
cd yolo-mlx
python3 -m venv .venv && source .venv/bin/activate
pip install -e .
pip install -e ".[segment]"
pip install -e ".[convert]"

# 2. Download a pretrained segmentation model and convert to MLX format
bash scripts/download_yolo26_models.sh          # downloads all .pt weights to models/
yolo-mlx converters convert models/yolo26n-seg.pt -o models/yolo26n-seg.npz --verify

# 3. Run segmentation
mkdir -p images
curl -fsSL -o images/bus.jpg https://ultralytics.com/images/bus.jpg
```

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

### Setup and Model Conversion for Segmentation Training

Source: https://github.com/thewebai/yolo-mlx/blob/main/README.md

Installs YOLO-MLX with segmentation and conversion dependencies, then downloads and converts a pretrained segmentation model for training.

```bash
# 1. Setup (if not done already)
cd yolo-mlx
python3 -m venv .venv && source .venv/bin/activate
pip install -e .
pip install -e ".[segment]"
pip install -e ".[convert]"

# 2. Download and convert a pretrained segmentation model as starting weights
bash scripts/download_yolo26_models.sh
yolo-mlx converters convert models/yolo26n-seg.pt -o models/yolo26n-seg.npz --verify
```

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

### Set Up Development Environment

Source: https://github.com/thewebai/yolo-mlx/blob/main/CONTRIBUTING.md

Installs the project and development dependencies in a virtual environment.

```bash
python3 -m venv .venv
source .venv/bin/activate
pip install -e .[dev]
```

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

### Install Weight Conversion Dependencies

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_SEGMENTATION.md

Installs packages necessary for converting model weights from PyTorch (.pt) to MLX (.npz) format. This is a one-time setup for conversion.

```bash
# Install weight conversion dependencies (needed once to convert .pt → .npz)
pip install -e "[convert]"
```

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

### Setup: Download MOT17 Dataset and Create Sample Video

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_TRACKING.md

Downloads the MOT17 dataset and creates a sample video file for tracking demonstrations.

```bash
# ── Setup: Download MOT17 & create sample video ──
bash scripts/download_mot17.sh
python scripts/create_sample_video.py
```

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

### Activate Environment and Setup

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_TRACKING.md

Commands to activate the project environment, download models, and convert them for use with YOLO-MLX.

```bash
# Activate environment
cd yolo-mlx
source .venv/bin/activate

# ── Setup: Download models & convert ──
bash scripts/download_yolo26_models.sh
yolo-mlx converters convert models/yolo26n.pt -o models/yolo26n.npz --verify
```

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

### Install Pre-commit Hooks

Source: https://github.com/thewebai/yolo-mlx/blob/main/CONTRIBUTING.md

Installs the pre-commit framework to automatically check code quality before commits.

```bash
make pre-commit-install
```

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

### Install Conversion Dependencies

Source: https://github.com/thewebai/yolo-mlx/blob/main/CONTRIBUTING.md

Installs additional dependencies required for weight conversion tasks.

```bash
pip install -e .[convert]
```

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

### Install Conversion Dependencies

Source: https://github.com/thewebai/yolo-mlx/blob/main/README.md

Installs dependencies required for converting model weights from .pt to .npz format.

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

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

### Install Optional Benchmark Dependencies

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_SEGMENTATION.md

Installs PyTorch and torchvision for running optional benchmarks comparing MLX performance against PyTorch MPS/CPU.

```bash
# (Optional) Install PyTorch MPS/CPU comparison benchmarks
pip install torchvision                           # optional, for MPS/CPU benchmarks
```

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

### Setup and Model Conversion for Inference

Source: https://github.com/thewebai/yolo-mlx/blob/main/README.md

Sets up the environment and converts a YOLOv26 model to MLX format for inference. Ensure you are in the yolo-mlx directory.

```bash
cd yolo-mlx
python3 -m venv .venv && source .venv/bin/activate
pip install -e .
pip install -e ".[convert]"
bash scripts/download_yolo26_models.sh          # downloads all .pt weights to models/
yolo-mlx converters convert models/yolo26n.pt -o models/yolo26n.npz --verify
mkdir -p images
curl -fsSL -o images/bus.jpg https://ultralytics.com/images/bus.jpg
```

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

### Install YOLO26 MLX and Dependencies

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_TRAINING_BENCHMARK.md

Install the yolo-mlx package and its core dependencies. Also install optional dependencies for weight conversion, COCO evaluation, and PyTorch comparison benchmarks.

```bash
# Install yolo-mlx and its core dependencies (mlx, numpy, pillow, pyyaml, tqdm)
pip install -e .

# Install weight conversion dependencies (needed once to convert .pt → .npz)
pip install -e ".[convert]"

# Install COCO evaluation and chart generation tools
pip install pycocotools matplotlib

# (Optional) Install PyTorch MPS/CPU comparison benchmarks
pip install torchvision                           # optional, for MPS/CPU benchmarks
```

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

### Install YOLO26 MLX Weight Conversion Dependencies

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_INFERENCE_VALIDATION.md

Installs additional dependencies required for converting PyTorch model weights (.pt) to MLX format (.npz). This is typically a one-time setup.

```bash
# Install weight conversion dependencies (needed once to convert .pt → .npz)
pip install -e ".[convert]"
```

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

### Install COCO Evaluation and Charting Tools

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_INFERENCE_VALIDATION.md

Installs tools necessary for COCO dataset evaluation (mAP) and generating performance charts.

```bash
# Install COCO evaluation and chart generation tools
pip install pycocotools matplotlib
```

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

### Install YOLO26 MLX and Dependencies

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_TRACKING.md

Install the yolo-mlx package and its core, tracking, and conversion dependencies using pip.

```bash
# Install yolo-mlx and its core dependencies (mlx, numpy, pillow, pyyaml, tqdm)
pip install -e .

# Install tracking dependencies (OpenCV for video I/O, lap for assignment)
pip install -e ".[tracking]"

# Install weight conversion dependencies (needed once to convert .pt → .npz)
pip install -e ".[convert]"
```

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

### Install Optional PyTorch Comparison Dependencies

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_INFERENCE_VALIDATION.md

Installs torchvision for optional benchmarking against PyTorch's MPS and CPU backends.

```bash
# (Optional) Install PyTorch MPS/CPU comparison benchmarks
pip install torchvision
```

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

### Manually Setup COCO Dataset Directories

Source: https://github.com/thewebai/yolo-mlx/blob/main/README.md

Manually creates directories and downloads necessary files for the COCO dataset if the automatic script is not used. This includes images, annotations, and labels.

```bash
mkdir -p datasets/coco/images datasets/coco/annotations datasets/coco/labels

curl -L -o datasets/coco/images/val2017.zip http://images.cocodataset.org/zips/val2017.zip
unzip datasets/coco/images/val2017.zip -d datasets/coco/images/
rm datasets/coco/images/val2017.zip

curl -L -o datasets/coco/annotations/annotations_trainval2017.zip http://images.cocodataset.org/annotations/annotations_trainval2017.zip
unzip datasets/coco/annotations/annotations_trainval2017.zip -d datasets/coco/
rm datasets/coco/annotations/annotations_trainval2017.zip

curl -L -o datasets/coco/labels/val2017.zip https://github.com/ultralytics/assets/releases/download/v0.0.0/coco2017labels-segments.zip
unzip datasets/coco/labels/val2017.zip -d datasets/coco/
rm datasets/coco/labels/val2017.zip
```

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

### Install YOLO26 MLX Package and Core Dependencies

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_INFERENCE_VALIDATION.md

Installs the yolo-mlx package and its essential runtime dependencies using pip.

```bash
# Install yolo-mlx and its core dependencies (mlx, numpy, pillow, pyyaml, tqdm)
pip install -e .
```

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

### CLI Tracking with Frame Skipping

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_TRACKING.md

Process only every Nth frame for faster tracking by setting the `--vid-stride` option. This example processes every second frame.

```bash
# Skip frames for faster processing (process every 2nd frame)
yolo-mlx track --model models/yolo26n.npz --source video.mp4 --vid-stride 2 --save
```

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

### Install Segmentation Dependencies

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_SEGMENTATION.md

Installs additional packages required for instance segmentation tasks, including COCO mAP evaluation and mask visualization.

```bash
# Install segmentation dependencies (pycocotools for COCO mAP, matplotlib for charts, opencv-python for masks)
pip install -e "[segment]"
```

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

### Install Segmentation Dependencies

Source: https://github.com/thewebai/yolo-mlx/blob/main/README.md

Installs the necessary dependencies for segmentation tasks, including pycocotools, matplotlib, and opencv-python. This is required for model.predict() with task="segment", COCO mask mAP, and chart generation.

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

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

### YOLO-seg Label Format Example

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_SEGMENTATION.md

Illustrates the polygon vertex annotation format for segmentation labels. Each line specifies a class ID followed by normalized polygon coordinates.

```text
class_id x1 y1 x2 y2 x3 y3 ... xN yN
```

```text
0 0.123 0.456 0.234 0.567 0.345 0.678 0.456 0.789 0.123 0.456
45 0.500 0.200 0.600 0.300 0.700 0.400 0.500 0.200
```

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

### Create and Activate Virtual Environment

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_TRAINING_BENCHMARK.md

Navigate to the project directory and create a Python virtual environment named .venv, then activate it.

```bash
cd yolo-mlx

# Create a new virtual environment
python3 -m venv .venv

# Activate it
source .venv/bin/activate
```

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

### Activate Environment and Run Inference Benchmark

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_SEGMENTATION.md

Quick reference to activate the project environment and run an inference benchmark for all models, skipping MPS and CPU backends.

```bash
# Activate environment
cd yolo-mlx
source .venv/bin/activate

# ── Inference benchmark (all models, MLX only) ──
python scripts/benchmark_yolo26_seg_inference.py --skip-mps --skip-cpu
```

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

### All Options Combined for MLX Benchmark

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_SEGMENTATION.md

Run the MLX YOLOv26-seg training benchmark with multiple custom options including models, epochs, and batch size. This demonstrates a comprehensive configuration.

```bash
python scripts/benchmark_yolo26_seg_training_mlx.py --models n s m l x --epochs 10 --batch 4
```

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

### Download MOT17 Dataset and Create Sample Video

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_TRACKING.md

Download the MOT17 dataset for evaluation and create a sample pedestrian video from its frames using provided scripts.

```bash
# Download MOT17 dataset (~5.5 GB) for evaluation and demo videos
bash scripts/download_mot17.sh

# Create a short sample pedestrian video (~3s, 1080p) from MOT17 frames
python scripts/create_sample_video.py
```

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

### CPU Training Benchmark (All Models)

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_TRAINING_BENCHMARK.md

Run the PyTorch CPU training benchmark for all specified models.

```bash
python scripts/benchmark_yolo26_training_cpu.py --models n s m l x --epochs 10 --batch 4
```

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

### Activate Environment and Run Inference Benchmark

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_INFERENCE_VALIDATION.md

Activate the project's virtual environment and run the inference benchmark script for MLX only.

```bash
cd yolo-mlx
source .venv/bin/activate
python scripts/benchmark_yolo26_inference.py --skip-mps --skip-cpu
```

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

### Benchmark YOLOv26-seg Training on PyTorch CPU (Full)

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_SEGMENTATION.md

Run full training benchmarks for all YOLOv26-seg models using the PyTorch CPU backend. Specify multiple models, epochs, and batch size.

```bash
# ── CPU training benchmark (all models) ──
python scripts/benchmark_yolo26_seg_training_cpu.py --models n s m l x --epochs 10 --batch 4
```

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

### MLX Training Benchmark (All Models)

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_TRAINING_BENCHMARK.md

Run the MLX training benchmark for all specified models.

```bash
python scripts/benchmark_yolo26_training_mlx.py --models n s m l x --epochs 10 --batch 4
```

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

### Webcam Tracking via CLI

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_TRACKING.md

Perform real-time object tracking from a webcam using the command-line interface. The `--show` flag displays the output.

```bash
# Webcam tracking
yolo-mlx track --model models/yolo26n.npz --source 0 --show
```

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

### Benchmark YOLOv26 Training (Specific Models and Settings)

Source: https://github.com/thewebai/yolo-mlx/blob/main/README.md

Runs training benchmarks for specific YOLOv26 models ('n', 's') with custom settings, including 10 epochs and a batch size of 4.

```bash
python scripts/benchmark_yolo26_training_mlx.py --models n s --epochs 10 --batch 4
```

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

### Collect and Chart Benchmark Results

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_TRAINING_BENCHMARK.md

Collect all benchmark results and generate comparison charts.

```bash
python scripts/benchmark_yolo26_collect_results.py
python scripts/benchmark_yolo26_generate_charts.py
```

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

### Tracking Demo Script: Webcam

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_TRACKING.md

Execute the tracking demo script for real-time webcam feed processing, enabling live display of tracking results.

```bash
# Webcam tracking
python scripts/track_demo.py --model models/yolo26n.npz --source 0 --show
```

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

### Benchmark Inference with Custom Warmup and Runs

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_INFERENCE_VALIDATION.md

Configure custom warmup runs and timed runs for the inference benchmark script.

```bash
python scripts/benchmark_yolo26_inference.py --warmup 5 --runs 15
```

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

### CLI Tracking with BoT-SORT Tracker

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_TRACKING.md

Utilize the BoT-SORT tracker instead of the default ByteTrack by specifying the tracker configuration file.

```bash
# Use BoT-SORT instead of ByteTrack
yolo-mlx track --model models/yolo26n.npz --source video.mp4 --tracker botsort.yaml --save
```

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

### Download Test Image for Inference

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_INFERENCE_VALIDATION.md

Downloads a sample image to be used for testing and benchmarking inference.

```bash
mkdir -p images
curl -L -o images/bus.jpg https://ultralytics.com/images/bus.jpg
```

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

### Benchmark YOLOv26-seg Training on PyTorch MPS (Full)

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_SEGMENTATION.md

Run full training benchmarks for all YOLOv26-seg models using the PyTorch MPS backend. Specify multiple models, epochs, and batch size.

```bash
# ── MPS training benchmark (all models) ──
python scripts/benchmark_yolo26_seg_training_mps.py --models n s m l x --epochs 10 --batch 4
```

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

### Run Common Development Commands

Source: https://github.com/thewebai/yolo-mlx/blob/main/CONTRIBUTING.md

Executes common development tasks such as formatting, linting, and testing.

```bash
make format
make lint
make test
make check
```

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

### Collect Benchmark Results and Generate Charts

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_SEGMENTATION.md

Combine results from multiple benchmark runs and generate comparison charts. Requires matplotlib. Supports outputting charts in PDF format and customizing DPI and output directory.

```bash
# Combine all JSON results into one file
python scripts/benchmark_yolo26_seg_collect_results.py

# Generate comparison charts
python scripts/benchmark_yolo26_seg_generate_charts.py

# PDF format (for publications)
python scripts/benchmark_yolo26_seg_generate_charts.py --format pdf

# Custom DPI and output directory
python scripts/benchmark_yolo26_seg_generate_charts.py --dpi 300 --output my_charts/
```

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

### Tracking Demo Script: Video File

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_TRACKING.md

Run the standalone tracking demo script for video file processing, with options to display results and save the annotated output.

```bash
# Video file tracking (display + save)
python scripts/track_demo.py --model models/yolo26n.npz --source images/pedestrians.mp4 --show --save
```

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

### CLI Video Tracking with Display and Save

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_TRACKING.md

Track objects in a video file using the CLI, displaying the results in real-time and saving the annotated video.

```bash
# Track with display
yolo-mlx track --model models/yolo26n.npz --source images/pedestrians.mp4 --show --save
```

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

### Run Segmentation Training Benchmarking with Custom Settings

Source: https://github.com/thewebai/yolo-mlx/blob/main/README.md

Benchmark segmentation training with specific models (e.g., 'n' and 's'), a custom number of epochs, and batch size. The output file can be specified using the --output flag.

```bash
python scripts/benchmark_yolo26_seg_training_mlx.py --models n s --epochs 10 --batch 4
```

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

### Configure MLX Training Benchmark Options

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_TRAINING_BENCHMARK.md

Customize the MLX training benchmark by specifying models, epochs, batch size, learning rate, and output path.

```bash
# Specific models only
python scripts/benchmark_yolo26_training_mlx.py --models n s

# Custom epochs and batch size
python scripts/benchmark_yolo26_training_mlx.py --epochs 5 --batch 2

# Custom learning rate
python scripts/benchmark_yolo26_training_mlx.py --lr 0.0001

# Custom output path
python scripts/benchmark_yolo26_training_mlx.py --output my_results.json

# All options combined
python scripts/benchmark_yolo26_training_mlx.py --models n s m l x --epochs 10 --batch 4
```

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

### Customize Sample Video Creation

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_TRACKING.md

Modify the sample video creation script to use a different MOT17 sequence or frame count, or specify a custom output path.

```bash
# Custom sequence and frame count
python scripts/create_sample_video.py --sequence MOT17-04-SDP --frames 60

# Custom output path
python scripts/create_sample_video.py --output my_video.mp4
```

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

### Basic Video Tracking via CLI

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_TRACKING.md

Track objects in a video file using the command-line interface. The `--save` flag stores the annotated video.

```bash
# Track a video file
yolo-mlx track --model models/yolo26n.npz --source images/pedestrians.mp4 --save
```

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

### Run All MLX Training Benchmarks

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_SEGMENTATION.md

Execute the MLX training benchmark script to train and evaluate all YOLOv26-seg models. The output is saved to a JSON file.

```bash
python scripts/benchmark_yolo26_seg_training_mlx.py
```

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

### Custom Epochs and Batch Size for MLX Benchmark

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_SEGMENTATION.md

Configure the number of training epochs and the batch size for the MLX YOLOv26-seg training benchmark. This allows for customized training runs.

```bash
python scripts/benchmark_yolo26_seg_training_mlx.py --epochs 5 --batch 2
```

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

### MPS Training Benchmark (All Models)

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_TRAINING_BENCHMARK.md

Run the PyTorch MPS training benchmark for all specified models.

```bash
python scripts/benchmark_yolo26_training_mps.py --models n s m l x --epochs 10 --batch 4
```

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

### Tracking Demo Script: Frame-by-Frame Mode

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_TRACKING.md

Run the tracking demo script in a custom frame-by-frame processing mode, allowing for detailed analysis and manipulation of each video frame.

```bash
# Frame-by-frame custom processing mode
python scripts/track_demo.py --model models/yolo26n.npz --source images/pedestrians.mp4 --mode framewise --show
```

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

### Collect and Generate Benchmark Results

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_TRAINING_BENCHMARK.md

Combine all JSON results into one file and generate comparison charts. Supports PDF format for publications.

```bash
# Combine all JSON results into one file
python scripts/benchmark_yolo26_collect_results.py

# Generate comparison charts
python scripts/benchmark_yolo26_generate_charts.py

# PDF format (for publications)
python scripts/benchmark_yolo26_generate_charts.py --format pdf
```

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

### Run MLX Training Benchmark

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_TRAINING_BENCHMARK.md

Execute the MLX training benchmark script. You can specify models, epochs, and batch size.

```bash
python scripts/benchmark_yolo26_training_mps.py
python scripts/benchmark_yolo26_training_mps.py --models n s --epochs 5 --batch 2
```

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

### Benchmark YOLOv26-seg Training on PyTorch CPU

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_SEGMENTATION.md

Execute training benchmarks for YOLOv26-seg models on the PyTorch CPU backend. Offers the same CLI flags as the MPS script, plus a --threads flag to control CPU thread count. This is the slowest backend, useful for baseline comparisons.

```bash
python scripts/benchmark_yolo26_seg_training_cpu.py
python scripts/benchmark_yolo26_seg_training_cpu.py --models n s --epochs 5 --batch 2

# Control CPU thread count
python scripts/benchmark_yolo26_seg_training_cpu.py --threads 4
```

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

### Run MLX Training Benchmark for All Models

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_TRAINING_BENCHMARK.md

Execute the MLX training benchmark script to train YOLO26 models using the pure MLX implementation and measure performance metrics.

```bash
python scripts/benchmark_yolo26_training_mlx.py
```

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

### Collect and Generate Tracking Benchmark Charts

Source: https://github.com/thewebai/yolo-mlx/blob/main/README.md

Collects tracking results into a single JSON file and then generates comparison charts for various metrics like MOTA, IDF1, and FPS.

```bash
python scripts/benchmark_tracking_collect_results.py
```

```bash
python scripts/benchmark_tracking_generate_charts.py
```

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

### Download YOLO26 Segmentation Models

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_SEGMENTATION.md

Downloads pre-trained PyTorch (.pt) weights for various YOLO26 segmentation models. These are required before conversion to MLX format.

```bash
# Use the download script (downloads both detection and segmentation models)
bash scripts/download_yolo26_models.sh

# Or download segmentation models manually
cd models
curl -L -o yolo26n-seg.pt https://github.com/ultralytics/assets/releases/download/v8.4.0/yolo26n-seg.pt
curl -L -o yolo26s-seg.pt https://github.com/ultralytics/assets/releases/download/v8.4.0/yolo26s-seg.pt
curl -L -o yolo26m-seg.pt https://github.com/ultralytics/assets/releases/download/v8.4.0/yolo26m-seg.pt
curl -L -o yolo26l-seg.pt https://github.com/ultralytics/assets/releases/download/v8.4.0/yolo26l-seg.pt
curl -L -o yolo26x-seg.pt https://github.com/ultralytics/assets/releases/download/v8.4.0/yolo26x-seg.pt
cd ..
```

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

### Generate Tracking Benchmark Charts

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_TRACKING.md

Creates comparison charts from the combined JSON results. Supports custom output formats and directories.

```bash
# Generate PNG charts (default)
python scripts/benchmark_tracking_generate_charts.py

# PDF format for publications
python scripts/benchmark_tracking_generate_charts.py --format pdf

# Custom output directory
python scripts/benchmark_tracking_generate_charts.py --output my_charts/
```

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

### Activate Environment and Run Quick COCO Validation

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_INFERENCE_VALIDATION.md

Activate the project's virtual environment and run a quick COCO validation test on a subset of images.

```bash
cd yolo-mlx
source .venv/bin/activate
python scripts/evaluate_coco_val.py --model yolo26n --data datasets/coco --subset 100
```

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

### Run Full MOT17 Benchmark with Shell Script

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_TRACKING.md

Execute an end-to-end benchmark for MOT17 using a shell script, which can run all models with the default ByteTrack tracker, or be configured for quick runs, specific trackers, or single models.

```bash
bash scripts/run_mot17_benchmark.sh
```

```bash
bash scripts/run_mot17_benchmark.sh --quick
```

```bash
bash scripts/run_mot17_benchmark.sh --tracker botsort
```

```bash
bash scripts/run_mot17_benchmark.sh --model yolo26s
```

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

### Evaluate MOT17 with PyTorch MPS Comparison

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_TRACKING.md

Compares YOLO-MLX performance against PyTorch MPS for all models on the MOT17 dataset.

```bash
# ── MOT17 evaluation (PyTorch MPS comparison) ──
python scripts/evaluate_mot17_pytorch.py --model all --device mps
```

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

### Track Video with YOLO-MLX (Python API)

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_TRACKING.md

Demonstrates how to perform object tracking on a video file using the YOLO-MLX Python API.

```python
# ── Track a video (Python) ──
python -c "
from yolo26mlx import YOLO
model = YOLO('models/yolo26n.npz')
model.track('images/pedestrians.mp4', conf=0.25, save=True)
"
```

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

### Benchmark YOLOv26-seg Training on PyTorch MPS

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_SEGMENTATION.md

Run training benchmarks for YOLOv26-seg models using the PyTorch MPS backend. Supports standard CLI flags for model selection, epochs, and batch size. Default learning rate is 0.00001.

```bash
python scripts/benchmark_yolo26_seg_training_mps.py
python scripts/benchmark_yolo26_seg_training_mps.py --models n s --epochs 5 --batch 2
```

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

### Run PyTorch CPU Training Benchmark

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_TRAINING_BENCHMARK.md

Execute the PyTorch CPU training benchmark script. Supports same CLI flags as MPS script, plus --threads to control CPU thread count.

```bash
python scripts/benchmark_yolo26_training_cpu.py
python scripts/benchmark_yolo26_training_cpu.py --models n s --epochs 5 --batch 2

# Control CPU thread count
python scripts/benchmark_yolo26_training_cpu.py --threads 4
```

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

### Benchmark YOLOv26-seg Training on MLX

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_SEGMENTATION.md

Run training benchmarks for YOLOv26-seg models using the MLX backend. Supports specifying multiple models, epochs, and batch size. Default learning rate is 0.000119.

```bash
# ── MLX training benchmark (all models) ──
python scripts/benchmark_yolo26_seg_training_mlx.py --models n s m l x --epochs 10 --batch 4
```

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

### Run MOT17 PyTorch Comparison Evaluation

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_TRACKING.md

Evaluate YOLO-MLX models on the MOT17 benchmark using PyTorch with either the MPS or CPU backend. This allows for direct comparison of performance across different hardware and frameworks.

```bash
python scripts/evaluate_mot17_pytorch.py --model all --device mps
```

```bash
python scripts/evaluate_mot17_pytorch.py --model all --device cpu
```

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

### Track Webcam Feed with YOLO-MLX (CLI)

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_TRACKING.md

Tracks objects in real-time from a webcam feed using the YOLO-MLX command-line interface.

```bash
# ── Track webcam (CLI) ──
yolo-mlx track --model models/yolo26n.npz --source 0 --show
```

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

### Webcam Tracking with YOLO-MLX

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_TRACKING.md

Real-time video tracking directly from a webcam. Press 'q' to quit the display window.

```python
# Real-time tracking from webcam (press 'q' to quit)
results = model.track(0, conf=0.25, show=True)
```

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

### Benchmark YOLO-26 Segmentation with Custom Warmup and Runs

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_SEGMENTATION.md

Configures custom numbers for warmup and timed runs during inference benchmarking.

```bash
python scripts/benchmark_yolo26_seg_inference.py --warmup 5 --runs 15
```

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

### Run Specific Models with MLX Benchmark

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_SEGMENTATION.md

Benchmark only specific YOLOv26-seg models (e.g., 'n' and 's') using the MLX training script. This allows for focused testing.

```bash
python scripts/benchmark_yolo26_seg_training_mlx.py --models n s
```

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

### Evaluate MOT17 with YOLO-MLX (All Models)

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_TRACKING.md

Evaluates all available YOLO-MLX models against the MOT17 dataset.

```bash
# ── MOT17 evaluation (all models) ──
python scripts/evaluate_mot17.py --model all
```

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

### Run Pre-commit Hooks on All Files

Source: https://github.com/thewebai/yolo-mlx/blob/main/CONTRIBUTING.md

Manually triggers the pre-commit hooks to run on all files in the repository.

```bash
make pre-commit-run
```

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

### Download Official YOLO26 PyTorch Models

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_TRAINING_BENCHMARK.md

Download the official YOLO26 pretrained weights (.pt files) from Ultralytics using a provided script or manually via curl.

```bash
# Use the download script
bash scripts/download_yolo26_models.sh

# Or download manually
cd models
curl -L -o yolo26n.pt https://github.com/ultralytics/assets/releases/download/v8.4.0/yolo26n.pt
curl -L -o yolo26s.pt https://github.com/ultralytics/assets/releases/download/v8.4.0/yolo26s.pt
curl -L -o yolo26m.pt https://github.com/ultralytics/assets/releases/download/v8.4.0/yolo26m.pt
curl -L -o yolo26l.pt https://github.com/ultralytics/assets/releases/download/v8.4.0/yolo26l.pt
curl -L -o yolo26x.pt https://github.com/ultralytics/assets/releases/download/v8.4.0/yolo26x.pt
cd ..
```

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

### Benchmark YOLOv26 Inference (Specific Models)

Source: https://github.com/thewebai/yolo-mlx/blob/main/README.md

Runs inference benchmarks for specific YOLOv26 models (e.g., 'n' and 's'), skipping MPS and CPU backends.

```bash
python scripts/benchmark_yolo26_inference.py --models n s --skip-mps --skip-cpu
```

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

### Run Full MOT17 Benchmark

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_TRACKING.md

Executes the complete MOT17 benchmark script, likely encompassing multiple evaluation steps.

```bash
# ── Full MOT17 benchmark ──
bash scripts/run_mot17_benchmark.sh
```

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

### Run MOT17 Evaluation with BoT-SORT Tracker

Source: https://github.com/thewebai/yolo-mlx/blob/main/README.md

Performs MOT17 evaluation using the BoT-SORT tracker instead of the default ByteTrack.

```bash
python scripts/evaluate_mot17.py --model yolo26n --tracker botsort
```

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

### Run Instance Segmentation with YOLO-MLX

Source: https://github.com/thewebai/yolo-mlx/blob/main/README.md

Loads a pretrained segmentation model and performs instance segmentation on an image, saving the annotated output.

```python
from yolo26mlx import YOLO

model = YOLO("models/yolo26n-seg.npz", task="segment")
results = model.predict("images/bus.jpg")
print(results[0])                    # detection + mask summary
results[0].save()                    # saves annotated image with mask overlays to results/
```

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

### Run All Models Inference Benchmark

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_INFERENCE_VALIDATION.md

Execute the inference benchmark script to test all available YOLO models across all backends.

```bash
python scripts/benchmark_yolo26_inference.py
```

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

### Benchmark YOLOv26 Inference (More Runs)

Source: https://github.com/thewebai/yolo-mlx/blob/main/README.md

Runs inference benchmarks for YOLOv26 models with an increased number of timed runs (20) for more stable results, skipping MPS and CPU backends.

```bash
python scripts/benchmark_yolo26_inference.py --runs 20 --skip-mps --skip-cpu
```

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

### Run All YOLO-26 Segmentation Inference Benchmarks

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_SEGMENTATION.md

Executes inference benchmarks for all YOLO-26 segmentation models across available backends (MLX GPU, PyTorch MPS, PyTorch CPU).

```bash
python scripts/benchmark_yolo26_seg_inference.py
```

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

### Convert PyTorch Weights to MLX Format

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_TRAINING_BENCHMARK.md

Convert downloaded PyTorch .pt model files to MLX .npz format using the yolo-mlx converters tool. The --verify flag checks weight shape correctness.

```bash
yolo-mlx converters --help
yolo-mlx converters convert models/yolo26n.pt -o models/yolo26n.npz --verify
```

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

### Benchmark Inference with MLX Only

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_INFERENCE_VALIDATION.md

Run the inference benchmark script, skipping PyTorch comparisons and focusing solely on MLX performance.

```bash
python scripts/benchmark_yolo26_inference.py --skip-mps --skip-cpu
```

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

### CLI Tracking with Larger Input Size

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_TRACKING.md

Use a larger input image size for potentially improved accuracy, especially for smaller objects. This requires a model trained for that size.

```bash
# Larger input size for better accuracy
yolo-mlx track --model models/yolo26l.npz --source video.mp4 --imgsz 1280 --save
```

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

### Download MOT17 Dataset

Source: https://github.com/thewebai/yolo-mlx/blob/main/README.md

This script downloads the MOT17 dataset, which is approximately 5.5 GB in size.

```bash
bash scripts/download_mot17.sh
```

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

### Track Video with YOLO-MLX (CLI)

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_TRACKING.md

Performs object tracking on a video file using the YOLO-MLX command-line interface.

```bash
# ── Track a video (CLI) ──
yolo-mlx track --model models/yolo26n.npz --source images/pedestrians.mp4 --save
```

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

### Custom Output Path for MLX Benchmark

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_SEGMENTATION.md

Specify a custom output file path for the results of the MLX YOLOv26-seg training benchmark. This helps in organizing benchmark outputs.

```bash
python scripts/benchmark_yolo26_seg_training_mlx.py --output my_results.json
```

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

### Evaluate MOT17 with YOLO-MLX (Single Model)

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_TRACKING.md

Evaluates a single YOLO-MLX model against the MOT17 dataset.

```bash
# ── MOT17 evaluation (MLX, single model) ──
python scripts/evaluate_mot17.py --model yolo26n
```

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

### Basic Video Tracking with YOLO-MLX

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_TRACKING.md

Perform basic video tracking on a file. The results are saved to a file. Access per-frame results including track IDs and bounding boxes.

```python
from yolo26mlx import YOLO

model = YOLO("models/yolo26n.npz")

# Track pedestrians — saves annotated output to results/pedestrians_tracked.mp4
results = model.track("images/pedestrians.mp4", conf=0.25, save=True)

# Access per-frame results
for r in results:
    if r.boxes.is_track:
        print(r.boxes.id)     # track IDs (persistent across frames)
        print(r.boxes.xyxy)   # bounding boxes
```

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

### Activate Environment and Run Full COCO Validation

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_INFERENCE_VALIDATION.md

Activate the project's virtual environment and run the full COCO validation script for all models.

```bash
cd yolo-mlx
source .venv/bin/activate
python scripts/evaluate_coco_val.py --model all --data datasets/coco
```

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

### Custom Learning Rate for MLX Benchmark

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_SEGMENTATION.md

Set a specific learning rate for the MLX YOLOv26-seg training benchmark. This is useful for hyperparameter tuning.

```bash
python scripts/benchmark_yolo26_seg_training_mlx.py --lr 0.0001
```

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

### Run Multi-Object Tracking on Video

Source: https://github.com/thewebai/yolo-mlx/blob/main/README.md

Performs multi-object tracking on a video file using a YOLO-MLX model. Results are saved to a specified output file.

```python
from yolo26mlx import YOLO

model = YOLO("models/yolo26n.npz")

# Track pedestrians — saves annotated output to results/pedestrians_tracked.mp4
results = model.track("images/pedestrians.mp4", conf=0.25, save=True)
```

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

### Benchmark Inference with Custom Output Path

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_INFERENCE_VALIDATION.md

Specify a custom file path for saving the JSON output of the inference benchmark.

```bash
python scripts/benchmark_yolo26_inference.py --output my_results.json
```

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

### Benchmark YOLO-26 Segmentation with Custom Output Path

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_SEGMENTATION.md

Specifies a custom file path for saving the inference benchmark results.

```bash
python scripts/benchmark_yolo26_seg_inference.py --output my_results.json
```

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

### Collect YOLO-MLX Segmentation Benchmark Results

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_SEGMENTATION.md

Executes the script to collect benchmark results for YOLO-MLX segmentation models. Ensure the script path is correct.

```python
python scripts/benchmark_yolo26_seg_collect_results.py
```

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

### Access Segmentation Results

Source: https://github.com/thewebai/yolo-mlx/blob/main/README.md

Demonstrates how to access the detected bounding boxes and segmentation masks from the YOLO-MLX prediction results.

```python
boxes = results[0].boxes             # Boxes object — (N, 6) [x1, y1, x2, y2, conf, cls]
masks = results[0].masks             # Masks object — (N, H, W) binary masks
print(f"Detected {len(boxes)} objects with masks of shape {masks.data.shape}")
```

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

### Collect Tracking Benchmark Results

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_TRACKING.md

Aggregates per-model JSON files into a single combined JSON file for analysis. Supports specifying different trackers.

```bash
python scripts/benchmark_tracking_collect_results.py

# Collect botsort results
python scripts/benchmark_tracking_collect_results.py --tracker botsort
```

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

### Benchmark Inference with Specific Models

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_INFERENCE_VALIDATION.md

Run the inference benchmark script for only specified YOLO models (e.g., 'n' and 's').

```bash
python scripts/benchmark_yolo26_inference.py --models n s
```

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

### Benchmark YOLO-26 Segmentation on MLX Only

Source: https://github.com/thewebai/yolo-mlx/blob/main/GUIDE_SEGMENTATION.md

Runs inference benchmarks exclusively on the MLX backend, skipping PyTorch MPS and CPU comparisons.

```bash
python scripts/benchmark_yolo26_seg_inference.py --skip-mps --skip-cpu
```