Viso Suite
Evaluation Guide

Deep Learning Capabilities: Frameworks and AI Models

Table of Contents

Integrated AI Model Manager

Viso Suite allows teams to import, manage, version, and reuse AI models in one place.

  • Pre-Trained Algorithms
    Viso Suite provides a set of the most popular and best AI algorithms that have been trained on massive datasets. Those AI models can be used to build computer vision applications in the no-code editor. The AI vision experts at viso.ai are constantly scouting and integrating the best-performing algorithms and deep learning models.
  • Import Custom Algorithms
    Teams can import custom-trained or re-trained AI models using the AI model manager of Viso Suite. Custom algorithms can be used to recognize and detect special objects, events, or anomalies in real-time. The integrated data-collection abilities of Viso make it possible to achieve mission-critical accuracy with small datasets.
  • Deploy and swap AI models
    AI models can be used with modules, the building blocks in the visual editor, to develop and update applications rapidly. The AI model manager is closely integrated with the Visual Builder of Viso Suite, AI models can be set and exchanged with one single click. The ability to update the AI model and versions safely is a fundamental part of Viso’s future-proof architecture.

Deep Learning Frameworks and Libraries

Viso Suite supports all popular deep learning frameworks for computer vision, including the following:

  • TensorFlow 2.0 (TF). The popular open source TensorFlow machine learning library focused on neural networks. Originally developed by Google Brain.
  • TensorFlow Lite (TF Lite). TensorFlow Lite is a deep learning framework for on-device inference (Edge AI). It is optimized for ML on Edge Devices.
  • OpenVINO. OpenVINO tools optimize deep learning models for inference tasks on Intel hardware including CPU, integrated GPU, and Movidius VPU.
  • PyTorch. The most popular open source ML library PyTorch was developed by Facebook, it is based on the Torch library. PyTorch is used by Tesla and UBER to power computer vision solutions.
  • Chainer. Chainer is a deep learning framework widely used for the development and application of deep learning methods. It supports NVIDIA CUDA for high-performance applications.
  • OpenCV. OpenCV is the most popular computer vision library developed by Intel. It is aimed at real-time computer vision and features GPU acceleration.
  • OpenPose. OpenPose is a real-time multi-person detection library. It is one of the most popular open-source pose estimation technologies.
  • OpenPifPaf. OpenPifPaf is a neural network architecture for semantic keypoint detection (human body joints) and pose estimation at high speed.

ai model and framework manager in viso suite

Object Detection AI Models

Use the best performing pre-trained deep learning models and neural networks for object detection.

CPU
  • SSD Mobilenet v1, SSD Inception v2, SSDlite Mobilenet v2, SSD Mobilenet v2, SSD ResNet 50 v1
  • Faster RCNN ResNet50, Faster RCNN ResNet101, Faster RCNN Inception v2, Faster RCNN Inception ResNet v2, Faster RCNN ResNet 50 v1, Faster RCNN ResNet 101 v1
  • EfficientDet D0, EfficientDet D4
  • Yolov3, Yolov3 tiny
  • Yolov4, Yolov4 tiny
  • ResNet50
  • VGG19
VPU
  • Yolov3 FP16
  • Face Detection Retail 0004
  • SSD Mobilenet v2
  • Person Detection 0200, Person Detection 0201, Person Detection Retail 0013, Person Vehicle Bike Detection 2000, Person Vehicle Bike Detection Crossroad 0078, Vehicle Detection 0222, Vehicle Detection 0201
GPU
  • SSD MobileNet v1, SSD MobileNet v2, SSD Inception v2, SSDlite MobileNet v2, SSD ResNet50 v1
  • Faster RCNN Inception v2, Faster RCNN Inception ResNet v2, Faster RCNN ResNet50, Faster RCNN ResNet50 v1, Faster RCNN ResNet101, Faster RCNN ResNet101 v1
  • EfficientDet d0, EfficientDet d4
  • Yolov3, Yolov3 tiny
  • Yolov4, Yolov4 tiny
TPU
  • MobileNet SSD v1
  • MobileNet SSD v2

Image Classification AI Models

Robust pre-trained deep learning models and neural networks for image classification.

CPU
  • MobileNet
  • ResNet50
  • VGG19
VPU
  • Inception v1 (GoogLeNet)
GPU
  • MobileNet
  • ResNet50
  • VGG19
TPU
  • Inception v4
  • MobileNet

Keypoint Detection AI Models

Pre-trained deep learning models for keypoint detection, used for human pose estimation.

CPU
  • ResNet50, ResNet101, ResNet152, ResNext50
  • ShuffleNetV2x1, ShuffleNetV2x2
VPU
  • OpenPose
  • PoseNet MobileNet v1
GPU
  • ResNet 101
TPU
  • PoseNet
  • ResNet50

Object Tracking Algorithms

The object tracking algorithms to track multiple objects detected by the object detection module.

Tracking Algorithms
  • DLIB
  • MOSSE
  • CSRT
Re-Identification Algorithms
  • DeepSORT
  • Geo Distance

Object Segmentation AI Models

Popular object segmentation deep learning models for object segmentation tasks.

CPU
  • Mask RCNN Inception ResNet v2, Mask RCNN Inception v2
  • Mask RCNN ResNet101, Mask RCNN ResNet50
GPU
  • Mask RCNN Inception ResNet v2, Mask RCNN Inception v2
  • Mask RCNN ResNet101, Mask RCNN ResNet50

Face Recognition AI Models

Pre-trained deep learning models and algorithms for face recognition and facial attribute analysis.

Face Recognition Algorithms
  • VGG-Face
  • Google FaceNet
  • OpenFace
  • Facebook DeepFace
  • DeepID
  • Dlib
  • ArcFace
Face Detection Algorithms
  • OpenCV
  • Dlib
  • SSD
  • MTCNN
  • RetinaFace

Extending Viso Suite

Viso Suite is extensible with a platform, environment, tools, and features that enable organizations to extend functionality with custom code and integrate with other systems.