MTP PRODUCT
MATA
Model-Agnostic Task Architecture — a flexible framework for building AI Vision solutions.
What Is MATA?
MATA (Model-Agnostic Task Architecture) is an open-source Python framework that lets you write your vision pipeline once and swap any model — HuggingFace, ONNX, Torchvision — without changing a line of code.
Universal Model Loading
Load any model by HuggingFace ID, local ONNX file, TorchScript archive, or config alias with a single API call. The adapter layer auto-detects the runtime and handles the wiring for you.
Composable Graph Pipelines
Wire Detection → Segmentation → Embedding into typed directed acyclic graphs with parallel execution, conditional branching, and control flow. Run independent tasks in parallel for 1.5–3× speedup.
Zero-Shot Everything
CLIP for classification, GroundingDINO for detection, SAM for segmentation — no training required. Describe what you want in natural language and get results immediately.
Core Capabilities
MATA supports every major computer vision task through a consistent, minimal API. The same three-line pattern works across all task types.
Object Detection
Detect objects with bounding boxes and confidence scores using DETR, RT-DETR, GroundingDINO, YOLO, and more.
Segmentation
Instance, panoptic, and semantic segmentation with Mask2Former, SAM, SAM3 — including zero-shot promptable segmentation.
Classification
Standard and zero-shot image classification with ResNet, ViT, ConvNeXt, CLIP, and any HuggingFace classifier.
Depth Estimation
Monocular depth estimation with Depth Anything V1/V2 — produce per-pixel depth maps from a single image.
Object Tracking
Multi-object tracking across video streams with BotSort and ByteTrack, plus ReID for appearance-based re-identification.
OCR
Optical character recognition via EasyOCR, PaddleOCR, Tesseract, TrOCR, GOT-OCR2, and GLM-OCR — with bounding box support.
Vision-Language Models
Image captioning, visual question answering, and agent-mode tool calling with Qwen3-VL, Florence-2, Gemma 4, LLaVA-NeXT, and more.
Barcode & QR
Decode barcodes and QR codes from images using pyzbar or zxing-cpp — integrated as a first-class task.
Feature Embedding
Extract image feature vectors for similarity search, gallery matching, and recognition pipelines using CLIP, OSNet, and X-CLIP.
Graph Pipelines
MATA's unique differentiator is its typed task graph system. Compose tasks into validated DAGs that execute in parallel where possible — no manual orchestration required.
Typed DAGs
Every node declares its input and output artifact types. The graph validator catches wiring errors before runtime — no more shape mismatches at inference time.
Parallel Execution
Independent tasks (e.g., Detect + Classify + Depth) run simultaneously. The scheduler handles dependency resolution and deterministically merges results.
Conditional Control Flow
Branch on detection confidence, skip empty frames, or loop until a quality threshold is met — EarlyExit and While primitives handle it inline.
Mnemonic DSL
Graphs read like sentences: Detect → Filter → PromptBoxes → RefineMask → Fuse. Pipe style or explicit wiring — both compile to the same validated DAG.
Industry Applications
Manufacturing
Zero-shot defect detection, VLM assembly verification, per-component quality inspection.
Retail
Shelf product detection, brand classification, zero-shot product search, multi-modal stock assessment.
Autonomous Driving
Distance estimation, 4-way road scene analysis, real-time traffic tracking with VLM hazard reasoning.
Security & Surveillance
Crowd monitoring with person tracking, suspicious object detection with SAM + VLM reasoning.
Agriculture
Crop disease classification, aerial field segmentation with depth mapping, pest detection.
Healthcare
ROI segmentation for medical imaging, VLM-assisted report generation, conditional triage workflows.
Open Source
MATA is Apache 2.0 licensed and available on GitHub. It ships with a CLI, Jupyter notebook helpers, comprehensive documentation, and 5,500+ passing tests. Install with a single pip command and run your first detection in under a minute.
Interested in MATA?
Get in touch to learn how MATA can be integrated into your operational environment — or visit the GitHub repo to try it yourself.