Integrating Pose Estimation into a Mobile App

While developing a fitness app, we found that the raw PoseNet model produced false positives on every second frame—insufficient temporal stability. The solution was to implement pose tracking based on MediaPipe with custom OneEuroFilter smoothing. We integrate body keypoint detection into mobile apps—real-time identification of skeleton landmarks. This is in demand in fitness (rep counting, technique analysis), medicine (rehabilitation, gait diagnosis), and AR. Technically, the task is more complex than object detection: it requires not only accurate localization but also temporal stability. Our team has 5+ years of experience in computer vision and has delivered over 30 projects with pose analysis. Cost is determined after analysis and depends on the number of exercises and platforms. Time savings compared to in-house development can be as high as 40%. Typical engagements start at $3,000.

Choosing a Model: MoveNet, BlazePose, or ML Kit?

MoveNet Lightning is the best balance for mobile: TFLite-optimized, available via PoseLandmarker. BlazePose provides z-coordinates for 3D angles. ML Kit is simple but slightly less accurate. In our case, MoveNet Lightning was 2x faster than BlazePose on iPhone 11 (35 FPS vs 17 FPS), critical for real-time feedback.

Correct Rep Counting Approach

The naive approach—tracking the hip’s Y-coordinate—breaks in reality. The correct solution is to compute the knee joint angle via the scalar product of vectors [HIP → KNEE] and [KNEE → ANKLE]. A squat is an angle below 120°, standing up above 160°. State machine: STANDING → DOWN → STANDING = 1 rep. Angles using z-coordinates are more stable if the camera is not strictly from the side. Landmark smoothing is mandatory: raw data jumps 3–5 pixels between frames. Additionally, we use a median smoothing filter to clip outliers.

Best Smoothing Methods

For fitness, EMA is enough; for rehabilitation, OneEuroFilter. In our projects, we customize OneEuroFilter parameters per exercise: for squats low cutoff (min_cutoff=0.5), for fast movements higher (min_cutoff=1.0).

Integration: iOS and Android—Step by Step

1. Select model based on required accuracy and speed. MoveNet Lightning to start.
2. Prepare stack: MediaPipe Tasks Vision (SPM or Gradle) version 0.10.0.
3. Configure inference: PoseLandmarker with runningMode = .liveStream and PoseLandmarkerOptions.
4. Process results: normalized points (0..1) or worldLandmarks (in meters). For angles use worldLandmarks—they are independent of camera crop.
5. Render skeleton: on iOS—CAShapeLayer with animation, on Android—Canvas.drawLine on SurfaceView or Compose Canvas.
6. Test—check FPS and accuracy on target devices (iPhone 11, Pixel 6, Galaxy S21).

Common implementation mistakes:

• Drawing the skeleton in normalized coordinates without transforming to preview coordinates—forgetting aspect ratio and crop. Use convertNormalizedLandmarksToImageCoordinates.
• Running inference on the main thread—drops FPS to zero. Use background queues (userInteractive on iOS, SingleThreadExecutor on Android).
• No landmark smoothing—visual jitter and incorrect rep counting. Even simple EMA (α=0.7) solves it.

What Is Included in the Work

• Consultation on model selection and feasibility assessment.
• Architecture design (inference, UI, history storage).
• Integration of body keypoint detection on iOS and/or Android.
• Implementation of repetition counting or movement analysis logic.
• Custom smoothing and optimization for specific devices.
• Deliverables: complete source code for the pose module, detailed documentation for usage and calibration, access to code repositories, training for your development team.
• Post-launch support (bug fixes, updates, guarantee stability).

Timeline

Basic skeleton on video stream + one exercise: 1–2 weeks. Full fitness module with multiple exercises, voice feedback, and history: 3–4 weeks. Cost starts from $2,000 for basic integration and goes up to $8,000 for a full module with multiple exercises.

Contact us for a project assessment—we will help you choose a model and design a turnkey solution. Get a consultation on pose estimation integration today.

Machine Learning in Mobile Apps: CoreML, TFLite, and On-Device Models

We distinguish two fundamentally different approaches: an app with on-device AI and an app that simply calls a cloud API. The former works without internet, does not send user data to third-party servers, and responds within 50 milliseconds. The latter depends on network latency and pricing plans. Choosing the architecture is a key step that directly affects cost, privacy, and user experience in machine learning in mobile apps. Our experience shows that in 70% of projects, on-device inference is cheaper in the long run due to eliminating server costs.

How to Choose Between CoreML and TFLite for On-Device Inference?

CoreML — Apple's native framework for running ML models on device. Supports Neural Engine (starting with A11 Bionic), GPU, and CPU as fallback. Models are converted to .mlmodel format via coremltools from PyTorch, ONNX, or TensorFlow. Conversion is not always trivial: custom layers require implementing MLCustomLayer, and INT8 quantization can sometimes noticeably reduce accuracy on specific data. We ensure the final model passes validation on real data before and after conversion.

TensorFlow Lite — cross-platform alternative for Android and Flutter. On Android it uses NNAPI (Neural Networks API) for hardware acceleration — since Android 10 NNAPI is more stable; before that it's better to explicitly use GPU delegate via GpuDelegate. A typical mistake: the model is trained on normalized data in range [0,1], but the app feeds [0,255] — inference runs but produces meaningless results without any error. We include an automatic input data validation module in the SDK.

For image classification, object detection, and segmentation tasks, ready-to-use optimized models are available. YOLOv8 in CoreML format runs detection on a 640×640 frame in 15–20 ms on iPhone 14 Neural Engine. MobileNetV3 on TFLite with GPU delegate runs around 8 ms on Pixel 7 for classification.

What If You Need Text Generation On-Device?

Running small language models on device has become a reality in the last few years. Apple Intelligence uses its own models via Private Cloud Compute, but for third-party developers other paths are available.

llama.cpp with Metal backend on iOS is a working approach for phi-3-mini (3.8B parameters, 4-bit quantization, ~2.3 GB). Inference: 15–25 tokens/second on iPhone 15 Pro. For integration in Swift, use the Swift Package llama.swift or a wrapper via C interface llama.h. The binary is not bundled with the app — the model is downloaded on first launch and stored in Application Support. Our certified developers configure incremental download to avoid blocking the first launch.

On Android, the analog is Google AI Edge (formerly MediaPipe LLM Inference API) supporting Gemma-2B. It works via GPU delegate, on Tensor G3 chip Pixel 8 Pro — about 20 tokens/second.

Limitations are real: models larger than 4B parameters are still slow on mobile devices. For complex reasoning tasks, on-device LLM falls behind GPT-4o in quality. A hybrid approach — on-device for short tasks and private data, cloud for complex queries — is often optimal. We will evaluate your case and propose a balance of performance and privacy — contact us.

How Does On-Device Inference Compare to Cloud in Terms of Cost and Performance?

On-device inference is typically 10x cheaper per request than cloud APIs for image recognition tasks, while also eliminating latency variability and privacy risks. The table below summarizes the trade-offs.

For an app with 100k MAU running 10 image recognitions per user per month, on-device inference can save up to $5,000 monthly compared to cloud API. Get a free consultation on your ML architecture today.

Integrating OpenAI API and Other Cloud Models

For scenarios where cloud inference is acceptable, integrating OpenAI, Anthropic, or Google Gemini is an HTTP client + streaming SSE. In Swift, AsyncThrowingStream is convenient for streaming responses. In Kotlin, use Flow.

Critically: API keys must never be stored in the app bundle. Even an obfuscated key can be extracted from the IPA in 10 minutes using strings or frida. Correct architecture: mobile app → your own backend → OpenAI API. The backend controls rate limiting, logs requests, and protects the key.

What Is Included in the Work (Deliverables)

• Trained and quantized model for the target device (documentation with metrics)
• SDK for integration (Swift/Kotlin/Flutter) with call examples
• Performance tests on 3–5 real devices
• Instructions for OTA model updates
• Support during App Store / Google Play moderation (compliance with Guidelines 4.2, 5.1)
• 2 weeks of technical support after release

Typical Project Pipeline

1. Task analysis — measure latency, privacy, size, supported devices.
2. Model prototyping — in Python, evaluate accuracy on target data.
3. Conversion and quantization — for CoreML/TFLite with validation.
4. Integration into the app — model wrapped in a service layer (easy to swap CoreML ↔ TFLite ↔ cloud).
5. Testing — on real devices, measure FPS, RAM, battery.
6. Deployment — via TestFlight / Firebase App Distribution, monitor metrics.

Timelines: integration of a ready CoreML/TFLite model — 1–2 weeks, development of a custom model with mobile optimization — from 6 weeks, on-device LLM chat with personalization — 4–8 weeks.

Why We Take on Complex Cases?

10+ years of experience in mobile development, 50+ implemented AI/ML solutions, guarantee of compatibility with current iOS and Android versions. All projects undergo code review and load testing. The cost includes preparation of moderation documentation and training of your team.

Contact us — we will help you choose the architecture and implement ML in your app turnkey. Order an audit of your existing solution — we will assess the potential for server cost savings free of charge. In some projects, savings can reach significant amounts per month.