While several datasets for autonomous navigation have become available in recent years, they tended to focus on structured driving environments. This usually corresponds well-delineated infrastructure such as lanes, a small number of well-defined categories traffic participants, low variation object or background appearance and strong adherence rules. We propose DS, novel dataset road scene understanding unstructured environments where the above assumptions are largely not satisfied. It...

Hand pose recognition has been a problem of great interest to the Computer Vision and Human Interaction community for many years current solutions either require additional accessories at user end or enormous computation time. These limitations arise mainly due high dexterity human hand occlusions created in limited view camera. This work utilizes depth information novel algorithm recognize scale rotation invariant poses dynamically. We have designed volumetric shape descriptor enfolding...

Recent efforts in multi-domain learning for semantic segmentation attempt to learn multiple geographical datasets a universal, joint model. A simple fine-tuning experiment performed sequentially on three popular road scene demonstrates that existing frameworks fail at incrementally series of visually disparate domains. When new domain, the model catastrophically forgets previously learned knowledge. In this work, we pose problem incremental segmentation. Given trained particular goal is (i)...

Climate change presents significant challenges to ecosystems, economies, and societies globally. In Malaysia, a tropical country highly dependent on its natural resources, the impacts are evident in altered rainfall patterns, rising temperatures, extreme weather events. Despite these challenges, many studies still predominantly rely traditional statistical methods, which limit their capacity for making accurate climate predictions developing effective policy solutions.This study effectively...

Aim/Background: Plant diseases are a significant threat to global food security, affecting crop yields and economic stability. Early detection intervention crucial mitigating these losses. Currently, AI-driven plant disease largely depends on Convolutional Neural Networks (CNNs). However, the necessity for large datasets substantial computational resources limits their application in resource-constrained environments. This research investigates potential of fuzzy logic as an alternative...

In many Asian countries with unconstrained road traffic conditions, driving violations such as not wearing helmets and triple-riding are a significant source of fatalities involving motorcycles. Identifying penalizing riders is vital in curbing accidents improving citizens' safety. With this motivation, we propose an approach for detecting, tracking, counting motorcycle riding videos taken from vehicle-mounted dashboard camera. We employ curriculum learning-based object detector to better...

Learning to drive faithfully in highly stochastic urban settings remains an open problem. To that end, we propose a Multi-task from Demonstration (MT-LfD) framework which uses supervised auxiliary task prediction guide the main of predicting driving commands. Our involves end-to-end trainable network for imitating expert demonstrator's The intermediately predicts visual affordances and action primitives through direct supervision provide aforementioned guidance. We demonstrate such joint...

Autonomous driving and assistance systems rely on annotated data from traffic road scenarios to model learn the various object relations in complex real-world scenarios. Preparation training of deploy-able deep learning architectures require models be suited different adapt situations. Currently, existing datasets, while large-scale, lack such diversities are geographically biased towards mainly developed cities. An unstructured layout found several developing countries as India poses a...

Advancements in adaptive object detection can lead to tremendous improvements applications like autonomous navigation, as they alleviate the distributional shifts along pipeline. Prior works adopt adversarial learning align image features at global and local levels, yet instance-specific misalignment persists. Also, remains challenging due visual diversity background scenes intricate combinations of objects. Motivated by structural importance, we aim attend prominent regions, overcoming...

Large scale image datasets are a growing trend in the field of machine learning.However, it is hard to quantitatively understand or specify how various compare each other -i.e., if one dataset more complex harder "learn" with respect deep-learning based network.In this work, we build series relatively computationally simple methods measure complexity dataset.Furthermore, present an approach demonstrate visualizations high dimensional data, order assist visual comparison datasets.We our...

Few-shot learning is a problem of high interest in the evolution deep learning. In this work, we consider few-shot object detection (FSOD) real-world, class-imbalanced scenario. For our experiments, utilize India Driving Dataset (IDD), as it includes class less-occurring road objects image dataset and hence provides setup suitable for We evaluate both metric-learning meta-learning based FSOD methods, two experimental settings: (i) representative (same-domain) splits from IDD, that evaluates...

Self-supervised methods have shown promising results in denoising and dehazing tasks, where the collection of paired dataset is challenging expensive. However, we find that these fail to remove rain streaks when applied for image deraining tasks. The method's poor performance due explicit assumptions: (i) distribution noise or haze uniform (ii) value a noisy hazy pixel independent its neighbors. rainy pixels are non-uniformly distributed, it not necessarily dependant on neighboring pixels....

Autonomous driving relies on deriving understanding of objects and scenes through images. These images are often captured by sensors in the visible spectrum. For improved detection capabilities we propose use thermal to augment vision an autonomous vehicle. In this paper, present our investigations fusion spectrum using a publicly available dataset, it analyze performance object recognition other known datasets. We comparison night time imagery qualitatively demonstrate that significantly...

Autonomous surface vehicles (ASVs) have the potential to operate for extended periods of time in coastal, estuarine, and riverine environments a variety scientific, environmental, military applications. However, these are often highly dynamic unstructured, present new challenges autonomous operations. Toward goal achieving truly long-term operations unstructured maritime environments, we approach create shoreline map with an ASV real-time by combining analysis images from single...

Video conferencing systems are designed to deliver a collaboration experience that is as close possible actually meeting in person. Current systems, however, do poor job of integrating video streams presenting the users with shared content. Real and virtual content unnaturally separated, leading problems nonverbal communication overall conference experience. Methods interacting typically limited pointing mouse, which not natural component face-to-face human conversation. This paper presents...

Shoreline detection plays an important role in vision based navigation for autonomous surface vehicles (ASVs). It is a challenging task because of the diversity near-bank scenarios. In this paper, we present two-stage algorithm to find shoreline by employing multiple features. First, classify images into two types: reflection-unidentifiable and reflection-identifiable. Based on classification, are further analyzed with suitable techniques respectively. case, reflection subtle so water region...

Unconstrained Asian roads often involve poor infrastructure, affecting overall road safety. Missing traffic signs are a regular part of such roads. or non-existing object detection has been studied for locating missing curbs and estimating reasonable regions pedestrians on scene images. Such methods analyzing task-specific single cues. In this paper, we present the first most challenging video dataset objects, with multiple types which cues visible without in scenes. We refer to it as...

In this paper we describe Pixene, a photo sharing system that focuses on the capture and subsequent visualization consumption of interactions around shared photos, where may be with physically co-present friends family, or online one's social network. former scenario, richly multimodal involve pointing spoken comments. Remote interaction is primarily in form 'like's text comments networking sites. Pixene thus acts as common repository for over photos brings from co-located into single...

Incremental few-shot learning has emerged as a new and challenging area in deep learning, whose objective is to train models using very few samples of class data, none the old data. In this work we tackle problem batch incremental road object detection data from India Driving Dataset (IDD). Our approach, DualFusion, combines detectors manner that allows us learn detect rare objects with limited all without severely degrading performance de4tector on abundant classes. IDD OpenSet task,...