Contents
1. Summary of scene recognition technology
2. Supervised training of scene recognition technology
3. Summary of object recognition technology
4. Supervised learning of object detection technology
5. Galaxy camera uses
6. Reference
Summary of scene recognition technology
Scene recognition is a technique that uses AI deep learning models to distinguish specific objects or scenes in images. It uses supervised learning to train an AI model using object images or scenes you want to classify and the labels about them (classification results). Once trained, the AI model will be able to distinguish between the labels even if you put in images not used for training.
Supervised learning of scene recognition technology
Supervised learning is training an AI model with data that have correct answers. An image classification model is trained to categorize object and scene labels. Cifar10 [1] is the most widely used open dataset for image classification. But, training AI models with this sort of data requires an awful amount of it.
Once trained (updated with sufficient weightage), the AI model will be able to classify correct answers even if you put in an image that wasn't used in training.
Summary of object detection technology
Object detection technology uses AI deep learning models to detect labels (classification results) and boxes (regions) of specific objects in images. Scene recognition only classifies images, but object detection detects the object boxes and can tell their labels apart. Similarly, object detection uses supervised learning to train AI models. Once trained, the AI model can detect boxes and labels even for images not used for training.
Supervised learning of object detection technology
The image detection AI model is trained using the relevant answers for the boxes and labels of the objects you want to detect. Pascal VOC [2] is the most comprehensive open dataset used for image detection. Similar to scene recognition, AI models require a bulk of relevant image detection answers data to fully train.
A trained AI model (with sufficiently updated weights) identifies and circumscribes objects for detection, even if you include images that were not used in training.
Galaxy camera uses
Scene recognition technology
Galaxy Camera's scene recognition technology is trained to classify more than 30 types of objects or scenes using data acquired from real Galaxy cameras. The trained scene recognition technology can operate faster using Galaxy's AI-dedicated H/W chipset. Various features of the Galaxy camera use this technology.
1. Scene optimization feature
Recognizes over 30 types of scenes in Camera Preview and tunes the image quality optimized for those scenes.
2. Document scan feature
3. Single Take AI Time-Lapse effect
Object detection technology
Galaxy Camera's object detection technology is trained to detect more than 11 types of objects using data acquired from real Galaxy cameras. The trained object detection technology can run faster using Galaxy's AI-dedicated H/W chipset. The tech is used in tandem with different camera features.
1. Auto Focus feature enhancement
Used to detect specific objects in the camera preview to focus better on them.
2. Portrait mode feature enhancement
3. Night mode feature enhancement
4. Zoom Lock feature enhancement
CamCyclopedia Index - Samsung Members
You can also access CamCyclopedia anytime by going to Community -> Category (app) -> CamCyclopedia -> “CamCyclopedia Index”.
Reference
[1] Cifar10 Dataset : https://www.cs.toronto.edu/~kriz/cifar.html
[2] PascalVoc Dataset : http://host.robots.ox.ac.uk/pascal/VOC/
[3] Scene recognition technology is available in some Galaxy S/Note/Fold series models and in some Galaxy A/M series models.
[4] Object recognition technology is available in some Galaxy S/Fold series models.
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.