A new license plate localization algorithm is presented. Execution times of these operations can rather be long, especially where the image consists of large amount of either vehicle's linked components or the other existing objects. This algorithm combines the image processing techniques with some statistical methods and eventually a pattern checking method is also added. Here, minimum rectangle bounding box has been used instead of common bounding box methods, detaching essential details out of blobs and performance improvement, combined with a defined quantity called license plate possibility ratio (LPPR) and standard deviation, we present a robust method of license plate localization. New way of finding license plate's location out of so many rectangles, considering "Sensitive to angle" conditions for characters has also been presented, specifically. It should be noted that the proposed algorithm is regardless of plate's location. This paper presents a different approach on thresholding utilization called "Dynamic Thresholding" which would be obtained by orderly scan of various and sequential ranges of threshold values, confronting probable drawbacks of image lighting caused by lack of light and brightness or another light source radiation, in which, the most desirable threshold value for detection procedure is unknown. Pattern checking phase consists of "Character-Separator" system, using predefined libraries, allows us to detect and specialize state or the city where the license plate's pattern is getting utilized. Presenting the best percent accuracy (95.33%) among relevant well-known algorithms, and also the 25ms run time of the program, would be strong proofs of algorithm's efficiency.

No abstract

State of the art visual relation detection methods have been relying on features extracted from RGB images including objects' 2D positions. In this paper, we argue that the 3D positions of objects in space can provide additional valuable information about object relations. This information helps not only to detect spatial relations, such as standing behind, but also non-spatial relations, such as holding. Since 3D information of a scene is not easily accessible, we propose incorporating a pre-trained RGB-to-Depth model within visual relation detection frameworks. We discuss different feature extraction strategies from depth maps and show their critical role in relation detection. Our experiments confirm that the performance of state-of-the-art visual relation detection approaches can significantly be improved by utilizing depth map information.

A major challenge in scene graph classification is that the appearance of objects and relations can be significantly different from one image to another. Previous works have addressed this by relational reasoning over all objects in an image or incorporating prior knowledge into classification. Unlike previous works, we do not consider separate models for perception and prior knowledge. Instead, we take a multi-task learning approach by introducing schema representations and implementing the classification as an attention layer between image-based representations and the schemata. This allows for the prior knowledge to emerge and propagate within the perception model. By enforcing the model also to represent the prior, we achieve a strong inductive bias. We show that our model can accurately generate commonsense knowledge and that the iterative injection of this knowledge to scene representations, as a top-down mechanism, leads to significantly higher classification performance. Additionally, our model can be fine-tuned on external knowledge given as triples. When combined with self-supervised learning and with 1% of annotated images only, this gives more than 3% improvement in object classification, 26% in scene graph classification, and 36% in predicate prediction accuracy.

No abstract