A5006431582- Multimodal Machine Learning Applications
- Advanced Vision and Imaging
- Advanced Image and Video Retrieval Techniques
- Human Pose and Action Recognition
- Reinforcement Learning in Robotics
- Robotics and Sensor-Based Localization
- Domain Adaptation and Few-Shot Learning
- Video Analysis and Summarization
- Robot Manipulation and Learning
- Natural Language Processing Techniques
- Robotic Locomotion and Control
- Topic Modeling
- Image and Object Detection Techniques
- Image Processing Techniques and Applications
- Muscle activation and electromyography studies
- Video Surveillance and Tracking Methods
- Data Visualization and Analytics
- Image Processing and 3D Reconstruction
- vaccines and immunoinformatics approaches
- Cell Image Analysis Techniques
- Influenza Virus Research Studies
- Robotics and Automated Systems
- Artificial Immune Systems Applications
- Image Retrieval and Classification Techniques
- Robotic Path Planning Algorithms
Tehran University of Medical Sciences
2024-2025
Iranshahr University
2021-2025
Google (United States)
2024
DeepMind (United Kingdom)
2023-2024
Google (United Kingdom)
2024
University College London
2023
Mashhad University of Medical Sciences
2021
University of Washington
2014-2019
Seattle University
2015
Walt Disney (United States)
2015
Deep reinforcement learning has emerged as a promising and powerful technique for automatically acquiring control policies that can process raw sensory inputs, such images, perform complex behaviors. However, extending deep RL to real-world robotic tasks proven challenging, particularly in safety-critical domains autonomous flight, where trial-and-error is often impractical. In this paper, we explore the following question: train vision-based navigation entirely simulation, then transfer...
We investigated whether deep reinforcement learning (deep RL) is able to synthesize sophisticated and safe movement skills for a low-cost, miniature humanoid robot that can be composed into complex behavioral strategies. used RL train play simplified one-versus-one soccer game. The resulting agent exhibits robust dynamic skills, such as rapid fall recovery, walking, turning, kicking, it transitions between them in smooth efficient manner. It also learned anticipate ball movements block...
Humans are remarkably proficient at controlling their limbs and tools from a wide range of viewpoints. In robotics, this ability is referred to as visual servoing: moving tool or end-point desired location using primarily feedback. paper, we propose learning viewpoint invariant servoing skills in robot manipulation task. We train deep recurrent controller that can automatically determine which actions move the end-effector robotic arm object. This problem fundamentally ambiguous: under...
How can we know whether a statement about our world is valid. For example, given relationship between pair of entities e.g., `eat(horse, hay)', how this true or false in general. Gathering such knowledge and their relationships one the fundamental challenges extraction. Most previous works on extraction have focused purely text-driven reasoning for verifying relation phrases. In work, introduce problem visual verification phrases developed Visual Knowledge Extraction system called VisKE....
Large-scale recognition problems with thousands of classes pose a particular challenge because applying the classifier requires more computation as number grows. The label tree model integrates classification traversal so that complexity grows logarithmically. In this paper, we show how parameters can be found using maximum likelihood estimation. This new probabilistic learning technique produces significantly improved accuracy.
We present a system for applying sim2real approaches to "in the wild" scenes with realistic visuals, and policies which rely on active perception using RGB cameras. Given short video of static scene collected generic phone, we learn scene's contact geometry function novel view synthesis Neural Radiance Field (NeRF). augment NeRF rendering by overlaying other dynamic objects (e.g. robot's own body, ball). A simulation is then created engine in physics simulator computes dynamics from...
Humans are remarkably proficient at controlling their limbs and tools from a wide range of viewpoints angles, even in the presence optical distortions. In robotics, this ability is referred to as visual servoing: moving tool or end-point desired location using primarily feedback. paper, we study how viewpoint-invariant servoing skills can be learned automatically robotic manipulation scenario. To end, train deep recurrent controller that determine which actions move arm object. The problem...
We present MuSHR, the Multi-agent System for non-Holonomic Racing. MuSHR is a low-cost, open-source robotic racecar platform education and research, developed by Personal Robotics Lab in Paul G. Allen School of Computer Science & Engineering at University Washington. aspires to contribute towards democratizing field robotics as low-cost that can be built deployed following detailed, open documentation do-it-yourself tutorials. A set demos lab assignments Mobile Robots course Washington...
A scene category imposes tight distributions over the kind of objects that might appear in scene, appearance those and their layout. In this paper, we propose a method to learn structures can encode three main interlacing components scene: category, context-specific objects, Our experimental evaluations show our learned outperform state-of-the-art Deformable Part Models detecting scene. structure provides level understanding is amenable deep visual inferences. The also generate features...
Train in SimulationFig. 1: Domain Invariant Visual Servoing (DIVIS) learns collision-free goal reaching entirely simulation using dense multi-step rollouts and a recurrent fully convolutional neural network (bottom).DIViS can be directly deployed on real physical robots with RGB cameras for servoing to visually indicated goals as well semantic object categories (top).Abstract-Robots should understand both semantics physics functional the world.While robot platforms provide means interacting...
We investigate the use of prior knowledge human and animal movement to learn reusable locomotion skills for real legged robots. Our approach builds upon previous work on imitating or dog Motion Capture (MoCap) data a skill module. Once learned, this module can be reused complex downstream tasks. Importantly, due imposed by MoCap data, our does not require extensive reward engineering produce sensible natural looking behavior at time reuse. This makes it easy create well-regularized,...
We introduce Segment-Phrase Table (SPT), a large collection of bijective associations between textual phrases and their corresponding segmentations. Leveraging recent progress in object recognition natural language semantics, we show how can successfully build high-quality segment-phrase table using minimal human supervision. More importantly, demonstrate the unique value unleashed by this rich bimodal resource, for both vision as well understanding. First, that fine-grained labels...
We propose the problem of automated photo album creation from an unordered image collection. The is difficult as it involves a number complex perceptual tasks that facilitate selection and ordering photos to create compelling visual narrative. To help solve this problem, we collect (and will make available) new benchmark dataset based on Flickr images. Album Dataset provides variety annotations useful for task, including manually created albums various lengths. analyze provide experimental...
In this paper, we study the problem of answering visual analogy questions. These questions take form image A is to B as C what. Answering these entails discovering mapping from and then extending searching for D such that relation holds D. We pose learning an embedding encourages pairs analogous images with similar transformations be close together using convolutional neural networks a quadruple Siamese architecture. introduce dataset in natural images, show first results its kind on solving images.
Generalized Hough Transform (GHT) is an efficient method for detecting curves by exploiting the duality between points on a curve and parameters of that curve. However GHT has some practical limitations such as high computational cost huge memory requirement scaled rotated objects. In this paper new method, namely fuzzy generalized transform (FGHT), proposed alleviates these deficiencies utilizing concept inference system. FGHT R-table consists set rules which are fired gradient direction...
Steganography is a branch in the information hiding research area which aims to conceal data transmission between two parties. In this paper new method based on predictive coding proposed employs Quantization Index Modulation (QIM) for quantizing error values and embedding simultaneously. Furthermore, correction mechanism preserve histogram of cover image make it resistant against histogram-based attacks. To evaluate performance method, several experiments gray-level images are carried out...
Objectives: Considering the low rate of exclusive breastfeeding in mothers using drugs and role behavioral intention as one effective factors on breastfeeding, present study aimed to determine impact training based theory planned behavior continuation drug-dependent mothers. Materials Methods: This clinical trial was performed three hospitals Mashhad 2018. To this end, 52 drug-abusing were randomly divided into experimental control groups. The intervention protocol consisted four separate...