Photo Labelling API

The Photo Labelling API processes images of houses and generates the correct label for both indoor and outdoor images.

Process

At the outset, our sophisticated model discerns if an image represents an indoor or outdoor setting. If the model cannot conclusively ascertain this, the image is classified under the category 'other'.

Upon this preliminary determination, our system optimizes its subsequent analysis based on this initial classification. If the image is designated as 'indoor', our proficiency in room recognition comes into play, leveraging a specialized rooms model. In contrast, if the image is determined as 'outdoor', the advanced scene model is implemented.

Our approach ensures accurate categorization and refined analytics, providing an optimized user experience, and promoting the depth and precision of our work.

Scene Classification

When an input image is initially classified as 'outdoor', our system's subsequent task is to pinpoint the specific outdoor scene portrayed in the image. To accomplish this, our dedicated scene model is employed, delivering a scene class and an accompanying confidence percentage.

Interestingly, there's a unique scenario to consider! If the confidence percentage for the predicted image is equal to or below 70%, our system further provides the second most probable class for the image, along with its respective confidence percentage. This feature caters to situations where multiple classes may be identifiable within a single image.

To provide a comprehensive understanding, we have listed possible scene classes alongside corresponding example images below. This highlights our system's advanced ability to discern and categorize various outdoor scenes with notable precision.

Kindly be aware that the second most probable scene class is only disclosed if the confidence percentage of the primary scene class is 70% or less. This feature highlights the detailed accuracy of our classification system.

balcony-garden

street

view

no_building

facade

building_part

Example Prediction Scene Classification

The predicted image class is 'outdoor' with confidence 100%, the first scene class is 'balcony-garden' with confidence 36.92% and the second scene class is 'facade' with confidence 59.15%.

Rooms Classification

Once the input image is categorized as 'indoor', our sophisticated system proceeds to identify the specific type of room displayed in the image. This is achieved through our advanced rooms classification system. The result is not only the identified room class, but also a percentage indicating the confidence level of this prediction.

However, there's an intriguing twist! If the predicted room class is 'other room', our system delves deeper with an additional layer of classification. The outcome of this secondary analysis is the identification of a subclass, accompanied by a confidence percentage.

To provide a clearer understanding, we have included potential room classes or subclasses with corresponding example images below. This showcases our system's intricate capability to recognize and classify various room types with remarkable precision.

Please note, the Room Subclass is exclusively revealed if the identified room class is 'other_room'. This specificity is integral to our system's advanced classification process.

living_room

kitchen

bathroom

bedroom

other_room

hall_corridor

other_room

room (empty room)

other_room

scullery

Example Prediction Room Classification

The predicted image class is 'indoor' with 99.9% confidence, the room class is 'other_room' with confidence 96.18% and the room subclass is 'hall_corridor' with confidence 87.85%.