Visual Search for Clothing: AI Recognition and Indexing

Visual Search for Clothing: How It Works?

Imagine: a user photographs a random jacket on a passerby and within seconds sees a list of similar models with prices and purchase links. This is not magic — it's visual search, which we integrate into fashion apps. The task is not just to detect an object, but to find exact or similar items in the catalog, considering color, pattern, cut. We implement the full pipeline: from detection to retrieval. We'll assess your project in one day; turnkey delivery from two weeks. Contact us — we'll select the optimal architecture for your catalog.

Problems We Solve

Users often cannot describe in words what they see. This lowers conversion and increases returns. Visual search solves both: the user finds a product in seconds, and recommendation accuracy reduces the share of unsuitable purchases. According to the Google Visual Search Impact Report, implementing visual search boosts conversion by 15% and increases average order value by 12% through cross-sells.

Attribute Recognition

Two independent blocks: clothing attribute recognition and similar product search via image vector. Recognition accuracy reaches 95% when trained on specialized datasets. Server-side response time is under 200 ms.

Attribute recognition covers category (jacket, dress, sneakers), color, pattern (stripe, plaid, solid), style (casual, formal). Well-suited models: DeepFashion2 dataset, Fashionpedia annotations. Ready APIs: Google Vision AI (clothing detection), Clarifai Fashion Model, Snap ML Kit.

Similarity Search

The task is similarity search: image → embedding vector → nearest neighbor search in product base. Backbone — ViT (Vision Transformer) or ResNet50, fine-tuned on a fashion dataset. For vector search: Pinecone, Weaviate or pgvector if the catalog is up to 1–2 million items. With ViT, recall@1 is 5–7% higher than ResNet, but inference cost is 30% higher.

Why Segmentation of Multiple Items in a Frame Matters?

Often a frame contains a full outfit: jacket, jeans, sneakers. Segmentation allows searching each element separately. Without splitting, the search focuses on the largest object, ignoring the rest. Our models detect up to 10 objects per photo, each with confidence >0.85.

// iOS: pipeline from photo to search results
class FashionSearchService {

    func searchSimilar(image: UIImage) async throws -> FashionSearchResult {

        // 1. Clothing detection and crop
        let detectedItems = try await detectFashionItems(image: image)
        guard let primaryItem = detectedItems.first else {
            throw FashionError.noClothingDetected
        }

        // 2. Crop by bounding box
        let croppedImage = image.cropped(to: primaryItem.boundingBox)

        // 3. Parallel: attributes + embedding
        async let attributes = extractAttributes(croppedImage)
        async let embedding = generateEmbedding(croppedImage)

        // 4. Vector search via backend
        let (attrs, vec) = try await (attributes, embedding)
        let similarProducts = try await vectorSearch(
            embedding: vec,
            filters: SearchFilters(
                category: attrs.category,
                priceRange: nil   // price filter optional
            )
        )

        return FashionSearchResult(
            detectedItem: primaryItem,
            attributes: attrs,
            similarProducts: similarProducts
        )
    }
}

The user chooses what to search — by tapping on one of the detected outfit elements. This is better than automatically selecting the “largest object”.

How to Index the Product Catalog?

If you need to search your own store catalog, preliminary indexing is required. For each product card: image → embedding → record in vector store with metadata (SKU, price, category, color, stock status).

# Backend: indexing product catalog
async def index_product(product: Product, image_url: str):
    # Download and preprocessing
    image = await download_and_preprocess(image_url)

    # Generate embedding via fashion-specific model
    embedding = fashion_encoder.encode(image)  # numpy array [512]

    # Write to Pinecone
    await pinecone_index.upsert(vectors=[{
        "id": str(product.sku),
        "values": embedding.tolist(),
        "metadata": {
            "category": product.category,
            "color": product.color,
            "brand": product.brand,
            "price": product.price,
            "in_stock": product.in_stock,
            "image_url": product.thumbnail_url,
            "product_url": product.url
        }
    }])

Filtering by metadata during search (in_stock: true) is critical — showing “similar” items without stock is pointless. Typical conversion after implementing filtering: CTR increase of 20–30%.

Approach Comparison

Embedding Model Comparison

Process of Work

1. Analysis — discuss your catalog, target audience, metrics.
2. Design — choose stack, architecture, prepare prototype.
3. Implementation — write code, integrate model, configure index.
4. Testing — A/B tests, accuracy measurement, load testing.
5. Deployment — release to App Store and Google Play, monitoring.

What's Included

• Architectural documentation
• Source code SDK for iOS and Android
• Integration and operation manual
• Team training (2 days)
• 3 months warranty support

Timeline and Savings Estimates

Integration of a ready API — from 1 week. Full project with custom vector store — 1 to 2 months. Save up to 40% of development budget using ready-made components. Payback period — on average 3–6 months after launch.

We have implemented 15+ visual search projects for fashion retailers. Contact us for a project assessment — we'll choose the optimal solution. Get a consultation — we'll evaluate your project in one business day.

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.