Implementing Image Recognition in Mobile Apps

We implement turnkey image recognition: from camera capture to result display. With extensive experience, we've encountered all typical pitfalls — accuracy at 90% on test sets often drops to 70% on real photos. Below is how to avoid this and build a production-ready pipeline. Using ready-made pipelines (Core ML Vision, TFLite Task Library) cuts development time by 40% compared to custom implementation. We guarantee stable operation on all devices.

Image Sources and Their Peculiarities

Camera via AVCaptureSession (iOS) or CameraX (Android) is the most challenging case. Data arrives as CMSampleBuffer / ImageProxy in YUV_420_888 or BGRA format. Models expect RGB float32 or uint8. YUV → RGB conversion without native code introduces up to 40 ms delay. On Android we use ImageAnalysis.Builder().setOutputImageFormat(ImageAnalysis.OUTPUT_IMAGE_FORMAT_RGBA_8888) — this directly gives the needed format without manual conversion.

Gallery is simpler but has an EXIF orientation pitfall. UIImage on iOS correctly accounts for orientation when displaying, but the underlying CGImage may be rotated. Passing CGImage directly to the model degrades accuracy for vertically shot photos. The correct approach: CIImage(image: uiImage) → CIContext.createCGImage with applied orientation transformation.

On Android, BitmapFactory.decodeFile does not respect EXIF. Use ExifInterface with subsequent Matrix.postRotate. Otherwise the model receives a rotated image, reducing accuracy by 25%.

Why Accuracy Drops on Real Data?

Models are trained on datasets with specific distributions — user photos always have more variation (lighting, angle, background). Case study: a mushroom identification app. The EfficientNetV2-S model in Core ML achieved 91% on the test set but only 73% on user photos. Reason: the dataset was top-down shots; users shoot from below at an angle. Solution: added VNClassifyImageRequest with a confidence threshold of 0.6; when confidence is low, we prompt to reshoot with instructions. Accuracy rose to 84%.

How to Avoid Accuracy Loss During Format Conversion?

The key is to match preprocessing with the training pipeline. If the model was trained on ImageNet with normalization mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225], any deviation causes a 15–30% accuracy drop. Resize strictly as in training: if center_crop was used, do not use fit with padding. The model will see padding as part of the object and misclassify.

What Preprocessing Does an ML Model Need?

How We Build the Inference Pipeline

TFLite Task Library speeds integration by 2x compared to manual pipeline thanks to built-in preprocessing and model loading utilities.

Results arrive asynchronously in a callback — update UI on main thread: LiveData on Android, @MainActor on iOS. Inference takes 30–50 ms.

What's Included

1. Requirement audit: source, platform, target accuracy, latency.
2. Model and framework selection (Core ML / TFLite / ML Kit).
3. Implementation of preprocessing pipeline considering format and orientation.
4. Inference integration with asynchronous processing.
5. Testing on real user data — at least 100 samples.
6. Confidence threshold tuning.
7. Documentation and commented code.
8. Handover to CI/CD.

Timelines and Cost

Timelines: 1–2 weeks depending on model complexity and preprocessing readiness. Cost is calculated individually. We provide post-delivery support — bug fixes and consultations for up to 3 months free of charge.

let model = try VNCoreMLModel(for: EfficientNet().model)
let request = VNCoreMLRequest(model: model) { request, error in
    guard let results = request.results as? [VNClassificationObservation] else { return }
    // handle results
}
let handler = VNImageRequestHandler(cgImage: cgImage, orientation: .up)
try! handler.perform([request])

Contact us to evaluate your project — we'll analyze your case for free within a day. You can get a consultation through the form on our website. Order image recognition integration for your app 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.