Semantic segmentation, object detection, and image recognition. Computer vision applications integrated with deep learning provide advanced algorithms with deep learning accuracy. MATLAB® provides an environment to design, create, and integrate deep learning models with computer vision applications.
You can easily get started with specialized functionality for computer vision such as:
- Image and video labeling apps
- Image datastore to handle large amounts of data for training, testing, and validation
- Image and computer vision-specific preprocessing techniques
- Ability to import deep learning models from TensorFlow™-Keras and PyTorch for image recognition
Access and manage large amounts of data quickly and easily with ImageDatastore.
In deep learning, it’s all about having comprehensive data to build an accurate model. Data augmentation allows engineers to increase the number of samples and variations of samples to a training algorithm. Create more training images for robust classification by adding rotation and scale variance to your training images using image data augmentation techniques.
Labeling and Preprocessing
Image and video labeling, which includes pixel labeling and object region of interest, can save countless hours of manual labeling. Use image processing tools to crop, deblur, brighten, and enhance images before training a network.
Network Design, Training, and Evaluation
Interactively design networks, speed up training using NVIDIA® GPUs, and get to good results faster.
Import pretrained models using ONNX™, then use Deep Network Designer app to add, remove, or rearrange layers.
Whether you’re using one GPU, many GPUs, the cloud, or NVIDIA DGX, MATLAB supports multi-GPU training with one line of code.
Understand how your network performs at any point in time.
- Before training: Use network analyzer to analyze network layers and ensure layer input/output compatibility.
- During training: Visualize a plot of validation accuracy while the network trains and stop training at any time.
- After training: Simulate deep learning networks in Simulink with
control, signal processing, and sensor fusion components to assess the impact
of your deep learning model on system-level performance.
Deploy deep learning models anywhere - automatically generate code to run natively on ARM® and Intel® MKL-DNN. Import your deep learning models and generate CUDA® code, targeting TensorRT and CuDNN libraries.
Deep Learning for Computer Vision Examples
MATLAB provides tools for specific deep learning applications such as:
Visual Inspection and Defect Detection
Automated inspection and defect detection are critical for high-throughput quality control in production systems. With MATLAB, you can develop deep learning–based approaches to detect and localize different types of anomalies.
Semantic segmentation is the act of labeling each pixel into a category. This is a key technology for automated driving and medical image processing.
Object detection uses classification techniques like YOLO v2 and Faster-RCNN to identify objects in a scene.
Image and Video Classification
Identify objects in images and video using the latest research models and transfer learning techniques.
MATLAB makes processing 3D data possible with sparse and dense 3D techniques. Applications include Lidar classification and 3D stacks of medical images.