We’ll have a look at which computer vision platforms surveyed companies considered to be the best, as highlighted in our AI Software of Choice 2022 report.


Konduto

The company

Konduto used its e-Commerce, artificial intelligence, and payment gateways experience to create a platform using machine learning and browsing behavior tracking in order to fight online fraud. In 2019, it had 4,000 merchants in Latin America, 175 million processed orders, and prevented the loss of $820 million.

The platforms

This fraud analysis engine uses machine learning technology to prevent fraud in online shopping. By crossing the information from over 2,000 data points and consulting Konduto’s database, it lets buyers have a more confident shopping experience.

Each buyer’s unique behavior is analyzed, with the platform combing over searches, product views, product comparisons, and more, to monitor and identify fraudulent profiles.

Konduto Complete

Ideal for:

  • New businesses
  • e-Commerces
  • Focus on sales

Main features:

User behavior

Priority order

Review

Portal and reports

Manual review

Hacker behavior analysis through browsing.

One-click order prioritization.

Ask the company for a platform review.

Operation monitoring and reporting.

Konduto team manually checks suspect orders.

Konduto Performance

Ideal for:

  • Payment providers
  • Real-time decisions
  • High-volume retailers

Main features:

User behavior

Intuitive and practical portal

Multi-store units portal

Integrated bureaus

Rules management

Decision lists

Fingerprint

Hacker behavior analysis through browsing.

The risk and manual review team can analyze orders.

Manage risk operation in several stores.

Konduto is partnered with credit bureaus.

The algorithm creates automatic rules for analysis quality improvement.

Add or remove contacts.

Buyer’s device identity considered.

Rulex

The company

ATS Global is an automation, IT enterprise, and quality-focused company. Established in 1986, it has six business activities to support the digital transformation journey. It’s the Independent Solution Provider for smart digital transformation, with expertise in:

  • Advanced Planning and Scheduling
  • Application Lifecycle Management and CloudNXT
  • Automation and IT
  • Lean and Six Sigma
  • Manufacturing Execution System
  • Quality Management
  • Smart Manufacturing and Industry 4.0
  • Supply Chain Management

The platform

Rulex

As the first-ever digital decision management system that is powered by explainable AI, it combines data-derived knowledge and human expertise. Rulex has a full no-code enterprise-grade software that allows business users to interact with professional developers in optimizing, augmenting, and automating their decision-making processes.

Ideal for:

  • Enterprise-level

Main features:

Data ingestion

Data processing

Data visualization

Advanced analytics

Business rule management

Collect data from any source (JSON, MS Excel, cloud, database, etc.).

Grouping, merging, cleansing, pivoting, etc. of data without changing the code.

Graphical insights with easy dashboard creation.

Solve optimization issues, and perform advanced statistics. Create transparent clear box machine learning models.

Build your own business decision rules, test them, and combine them with data-driven rule models.

Detectron2

The company

Meta, Facebook’s new corporate name, has an organization that heads up the work on artificial intelligence: Meta AI. It focuses on the company’s own AI research, featuring research papers and open-source tools it develops. The social platform, Facebook, uses Meta AI’s neural network, image recognition, and more.

The platform

Detectron2

This open-sourced software system implements state-of-the-art object detection algorithms. Written in Python and powered by the Caffe2 deep learning framework, it aims to provide a high-performance and high-quality codebase for object detection research.

With Dectectron2, models can be exported to Caffe2 or TorchScript format for deployment. Compared to the first iteration, which had a throughput of 10 img/s, Detectron2 is much faster at 62 img/s. Available on GitHub.

Ideal for:

  • Beginners

Main features:

Object detection algorithm implementations

Modular and extensible design

New task support

New capabilities

Detectron2go

- All models from original Detectron

- Cascade R-CNN

- Panoptic FPN

- TensorMask


Users can plug custom module implementations into almost any section of an object detection system.

- Semantic segmentation

- Panoptic segmentation

- Panoptic segmentation

- Densepose

- Rotated bounding boxes

- PointRend

- DeepLab

- Etc.

An additional software layer for easier deployment of advanced new models to production.

MMDetection

The company

The OpenMMLab team wants to provide advanced research and development models and systems for the industry, alongside becoming a worldwide leader in open-source algorithm platforms for computer vision. Widely known across the globe, OpenMMLab has a wide variety of open-source projects for industrial applications and academic research, offering over 20 open datasets that are owned by MMLab.

The platform

MMDetection

This Python toolbox was built as a codebase just for object detection and instance segmentation. MMDetection is built in a modular way with PyTorch and is part of the OpenMMLab project. As an object detection toolbox and benchmark, it has inference and training codes and offers weight for more than addition. Available on GitHub.

Ideal for:

  • Quick training and inference

Main features:

Modular design

Support of multiple frameworks out of the box

Highly efficient

State-of-the-art 

Architectures

The detection framework is decomposed into different components, allowing for different modules to be combined.

- Faster RCNN

- Mask RCNN

RetinaNet

- Etc.

Basic box and mask operations run on GPUs.

Toolbox from codebase developed by the MMDet team (winners of COCO Detection Challenge 2018).

- Object detection

- instance segmentation

- Panoptic segmentation

- Etc.

Seldon Deploy

The company

Founded with the aim of accelerating the adoption of machine learning in order to solve some of the most challenging issues in the world, Seldon Technologies Ltd made its first open-source release in 2015. Having won awards in the industry, the company believes that machine learning is going to be at the center of every connected business soon.

The platform

Seldon Deploy

Offering governance and oversight for machine learning deployments, Seldon Deploy helps in deploying models in an audited way with Gitops. This enterprise product aims at accelerating deployment management on top of the open-source tools Seldon Alibi, KFserving, and Seldon Core.

You can:

  • Update models through Canary workflows.
  • Guarantee a safe model deployment with Gitops paradigm.
  • Audit model predictions with Black Model Explainers.
  • Monitor your running models and search response/request logs.
  • Deploy machine learning models effortlessly by using Seldon and KFServing open projects.

Ideal for:

  • Enterprise-level

Main features:

Deployment

Observability

Regulatory compliance

Model catalog

Model confidence

Stack stability

You can deploy models in your machine learning framework of choice.

Built-in dashboards let you monitor your running deployments.

A range of explainability methods helps to meet compliance and regulatory needs.

The centralized model catalog can connect models to data and training runs.

Model explainers help to adjust and understand features and anomaly detection flags drifts in data.

Full SLA, backward compatibility, and rolling updates alongside maintained integrations allow for a seamless install.

Google Cloud AI platform

The company

As a multinational technology company, Google focuses on search engine technology, artificial intelligence, computer software, cloud computing, quantum computing, and more. Some of Google’s innovations include Sycamore (quantum computing), Google Brain (transformer models), Sidewalk Labs (smart cities), and more.

The platform

The Google Cloud AI Platform is a suite of services targeted at building, deploying, and managing machine learning models in the cloud. It’s created for ease of use by data scientists and engineers, helping to streamline workflows. It offers data labeling, BigQuery Datasets, Notebooks, AutoML, training, AI explanations, What-If Tool, Vizier, prediction, and TensorFlow Enterprise services. GitHub repositories are available.

Ideal for:

  • Enterprise-level

Main features:

Tools to interact with AI platform

Collaborative ecosystem

Datasets

Machine learning products and services

- Google Cloud Console

- The Google Cloud CLI

- REST API

- Vertex AI Workbench user-managed notebooks

- Deep Learning VM

- TensorFlow

- Machine learning Kit

- CoLaboratory

- Google Open Source

- Objectron

- TimeDial

- AutoFlow

- DiscoFuse

- Open Images

- MAESTRO

- WikiSplit

- Conceptual Captions

- Etc.

- Cloud AI

- Cloud AutoML

- Cloud TPU (TPU v2, TPU v3, TPU v2 Pod, TPU v3 Pod)

PyTorch

The company

Mainly developed by Facebook’s AI Research lab, now known as Meta AI, PyTorch is an open-source machine learning framework for natural language processing, computer vision, and more. Tesla Autopilot’s deep learning software was built on top of PyTorch. Available on GitHub.

The platform

PyTorch was built to be both modular and flexible for research, offering support and stability for production deployment. While its mobile su[port is still experimental, it supports an end-to-end workflow on iOS and Android, from Python to development. PyTorch offers two high-level features:

  1. Deep neural networks are built on a tape-based automatic differentiation system.
  2. Tensor computing with a strong acceleration through GPU.

Ideal for:

  • Beginners

Main features:

Production-ready

Distributed training

Strong ecosystem

Cloud support

Native ONNX support

C++ frontend

TorchScript lets you seamlessly transition between graph and eager models. Speed up the path to production with TorchServe.

Performance optimization and scalable distributed training in both research and production - enabled by the torch.distributed backend.

Libraries and tools extend PyTorch, supporting development in natural language processing, computer vision, and more.

Frictionless development and easy scaling thanks to support on major cloud platforms.

Models can be exported in the standard OPen Neural Network Exchance (ONNX) format for direct access to ONNX-compatible platforms and more.

PyTorch has a pure C++ interface, which follows the architecture and design of the established Python frontend. Allows for research in low latency, high performance, and bare metal C++ applications

Python

The company

Developed by Guido van Rossum in the late 1980s, Python was created as the successor to the ABC programming language. It was first released in 1991 as Python 0.9.0, with future iterations added at later dates. It’s one of the most popular programming languages.

The platform

This high-level, general-purpose programming language highlights code readability with the utilization of significant indentation as its design philosophy. Python is garbage-collected (a form of automatic memory management) and dynamically typed, supporting a wide variety of programming paradigms like object-oriented and functional programming.

Ideal for:

  • Small and large-scale projects
  • Startups and enterprise-level
  • Experts and beginners

Main features:

Large standard library

Easy language

Object-oriented

Dynamically typed

Platform-independent

GUI support

Large community support

Including a variety of modules and packages, Python’s standard library means that you won’t have to write a lot from scratch.

Python is easy to learn thanks to its simple syntax. And it’s also easy to understand, making it easy to write.

It has object-oriented programming (OOP) concepts like polymorphism and inheritance.

Unlike Java, Python isn’t statically-typed. It follows duck-typing, meaning that “If it looks like a duck, swims like a duck, and quacks like a duck, it must be a duck.”

Programs written in Python can run on Windows, Mac, and Linux without needing to be written separately.

Easily create Graphical User Interfaces (GUI) by using tkinter, wxPthon, PyQt, or Pyside.

With one of the largest communities on Meetup and StackOverflow, it’s ideal when you need technical support.