Describing Objects by their Attributes

Describing Objects by their Attributes

In this research, attribute-centric approach has been developed. The object recognition permits the users to classify and identify the objects in an advance way besides the traditional naming task and offers many new aptitudes. Learning and recognizing the object attributes can be based on semantic approach but at the same time it cannot.  Semantic attributes define parts and style such as, nose shape as it is cylindrical or not.

The semantic attributes can be learned from interpretations and allow the users to describe objects and to categorize them. But the semantic attributes are not enough and appropriate for differentiating between object classes. For instance, it is hard to define and refer to the difference between dogs and cats besides having the visible and mark able visual dissimilarities. Therefore, the need to learn the non-semantic attributes that resemble to split in the pictorial features. These non-semantic features can be cultured by defining auxiliary tasks.  For example: to distinguish between cars and motorbikes using texture.

It has been established that deducing attributes of the objects is the main problem in recognition and describing the attributes of the objects. These attributes could be semantic qualities such as shapes and materials. Besides that these attributes may not be enough and adequate to classify and label the objects. This is the reason that the use of discriminative attributes is also necessary. Objects have and share different attributes. Thus, by means of predicted attributes as structures or feature the users can get a more solid and additional discriminative feature space acquiring and using both semantic and discriminative attributes open doors to some new and varied visual functions.

This system is also goaled to achieve and recognize the objects but in addition to that it can also describe them. For instance: besides identifying just cat it can also detect the Persian cat or the spotty cat etc.

Besides only focusing on distinct and identify the assignment, the system can inferred the attributes of the core problem of recognition. These attributes can be semantic and could be discriminative also.

Learning attributes presents a major new challenge: generalization across object categories, not just across instances within a category. In this research paper, we also introduce an original feature selection method for learning attributes that generalize well across categories. The research support its claims by thorough evaluation that provides insights into the limitations of the standard recognition paradigm of naming and demonstrates the new abilities provided by our attribute based framework.

Features and Attributes:

Semantic Attributes:

The three main types of semantic attributes has been used and utilized. Shape attributes mention to 2D and 3D properties such as “is 2D boxy”, “is 3D proxy”, “is cylindrical“, etc. Part attributes identify parts that are visible, such as “has head”, “has leg”, “has arm”, “has wheel”, “has wing”, “has window”. Material attributes describe what an object is made of, including “has wood”, “is furry”, “has glass”, “is shiny”.

 Discriminative Attributes:

Comprehensive and complete set of visual attributes has not yet been developed. This means that two alike and same classes of objects can share the same attributes.

Base Features:

There is broad variety of attributes which involves the representation of features to describe and evaluate other visual aspects. The bas features of the visual aspects include: color, materials, textures, words of the visuals, etc.

The advantages and the plus points of this system are that:

It provides the different and varied base for identifying and classifying the objects. It uses three different attributes to make the system and classification more clear and comprehensive.

We can not only identify and classify the objects using predicted attributes, characteristics and features, but can also describe unacquainted objects.

Additionally, these attribute classifiers can detect and announce the absence of some of the typical attributes of the objects, as well as occurrence of atypical attributes.

Finally, we can learn models for new object classes using few examples. Besides only focusing on distinct and identify the assignment, the system can inferred the attributes of the core problem of recognition. These attributes can be semantic and could be discriminative also.

We can even learn new categories with no visual examples, using textual descriptions instead.