Meta tags:
description=Jathushan Rajasegaran;
author=Jathushan Rajasegaran;

Headings (most frequently used words):

-

Text of the page (most frequently used words):
and (33), nbsp (16), jathushan (15), rajasegaran (15), the (14), arxiv (12), for (11), from (10), code (10), with (10), video (10), tracking (10), paper (9), computer (9), vision (9), recognition (9), tasks (9), project (9), page (9), jitendra (9), malik (9), models (9), poster (7), human (6), learning (6), representations (5), conference (5), neural (5), are (5), pattern (5), cvpr (5), khan (5), 2019 (4), world (4), study (4), capsule (4), networks (4), can (4), jayasekara (4), people (4), them (4), pose (4), angjoo (4), kanazawa (4), georgios (4), pavlakos (4), christoph (4), feichtenhofer (4), scaling (4), image (4), videos (4), may (4), international (3), suranga (3), seneviratne (3), that (3), ranga (3), new (3), information (3), processing (3), systems (3), neurips (3), salman (3), performing (3), appearance (3), location (3), predicting (3), reconstruction (3), demo (3), action (3), next (3), prediction (3), 2025 (3), training (3), also (3), worked (3), phd (3), 2026 (3), working (3), berkeley (3), was (3), counterfeit (2), apps (2), approach (2), google (2), writing (2), data (2), rodrigo (2), hirunima (2), sandaru (2), vinoj (2), jayasundara (2), increase (2), depth (2), skip (2), connections (2), oral (2), presentation (2), model (2), fahad (2), shahbaz (2), munawar (2), hayat (2), random (2), path (2), incremental (2), past (2), meta (2), monocular (2), using (2), their (2), 2023 (2), iccv (2), humans (2), humanoid (2), token (2), ilija (2), radosavovic (2), diffusion (2), time (2), compute (2), perceptual (2), flow (2), segmentation (2), rahul (2), ravishankar (2), properties (2), trained (2), masked (2), grok (2), imagine (2), xai (2), blog (2), thesis (2), research (2), deep (2), generative (2), ideas (2), into (2), before (2), completed (2), advised (2), prof (2), website, source, jon, barron, here, use, content, style, detect, playstore, wide, web, www, guillaume, jourjon, ashanie, gunathillake, naveen, karunanayake, multi, modal, embeddings, detecting, mobile, case, play, store, capture, actual, variations, present, hand, generate, more, retrain, winter, applications, wacv, textcaps, handwritten, character, very, small, datasets, cool, but, they, shallow, convolutions, deepcaps, going, deeper, width, resnet, like, adding, extra, when, introduced, ling, shao, selection, 2020, generic, easily, adapt, well, remember, old, slides, itaml, task, agnostic, 2021, lifting, then, 2022, results, best, finalist, top, recognize, atomic, actions, benefits, fully, transformerized, design, mesh, recovery, achieves, improved, precision, remarkable, robustness, shubham, goel, reconstructing, transformers, sarthak, kamat, 2024, real, control, problem, akin, word, language, koushil, sreenath, trevor, darrell, baifeng, shi, bike, zhang, locomotion, show, how, benefit, test, unify, such, estimation, optical, amodal, under, framework, translation, zeeshan, patel, autoencoder, gaussians, intermediate, this, allows, zero, shot, figure, ground, layering, edge, detection, while, same, supervised, finetuning, shiry, ginosar, rulilong, xinlei, chen, gaussian, autoencoders, llama, billion, parameters, trillion, visual, tokens, resulting, diverse, robotics, these, family, yossi, gandelsman, empirical, autoregressive, pre, multimodal, generation, system, creates, photoreal, images, short, native, audio, text, prompts, pretraining, lead, distillation, efforts, developed, post, algorithms, above, interests, lie, general, area, particularly, architectures, continual, three, hundred, year, least, modern, fermat, matching, classical, tamil, definition, seeing, surfaces, noisy, what, thiruvalluvar, knew, about, intelligence, toward, self, evolving, safeguards, demands, fire, trying, build, recently, started, turning, some, posts, mine, used, llms, help, materialize, have, not, spent, too, much, polishing, please, read, grain, salt, blogs, art, photos, github, scholar, bio, email, member, technical, staff, visiting, researcher, during, specifically, focusing, exploring, behaviours, terms, inference, understanding, coming, inception, institute, undergraduate, university, moratuwa, major, electronic, telecommunication, engineering, bachelor, rodigo, bair,

Text of the page (random words):
jathushan rajasegaran jathushan rajasegaran i am a member of technical staff at x ai working on video models before that i completed my phd at bair uc berkeley advised by prof jitendra malik i was also visiting researcher at meta ai working with dr christoph feichtenhofer during my phd at berkeley i worked on video models specifically focusing on human reconstruction tracking and recognition i also worked on exploring scaling behaviours of video generative models in terms of compute data and inference for understanding the world from videos before coming to berkeley i was working with prof salman khan at inception institute of ai i completed my undergraduate study at university of moratuwa with a major in electronic and telecommunication engineering my bachelor s thesis was advised by dr ranga rodigo email nbsp nbsp bio nbsp nbsp google scholar nbsp nbsp github nbsp nbsp photos nbsp nbsp art blogs i recently started turning some random ideas into blog posts the ideas are mine and i used llms to help materialize them into writing i have not spent too much time polishing them so please read them with a grain of salt the fire we re trying to build may 2026 toward self evolving ai and the safeguards it demands what thiruvalluvar knew about intelligence may 2026 a classical tamil definition for seeing past surfaces in a noisy world from fermat to flow matching may 2026 a three hundred year path from least action to modern generative models research my research interests lie in the general area of computer vision and deep learning particularly in learning from videos deep neural architectures and continual learning phd thesis may 2025 grok imagine xai project page blog demo grok imagine is xai s multimodal generation system that creates photoreal images and short videos with native audio from text prompts i worked on pretraining and lead distillation efforts for v0 9 and developed new post training algorithms for v1 and above an empirical study of autoregressive pre training from videos jathushan rajasegaran ilija radosavovic rahul ravishankar yossi gandelsman christoph feichtenhofer jitendra malik international conference on computer vision iccv 2025 nbsp project page arxiv code we trained llama models up to 1 billion parameters on 1 trillion visual tokens the resulting model can do diverse tasks from image video recognition video tracking action prediction and robotics we also study the scaling properties of these family of models gaussian masked autoencoders jathushan rajasegaran xinlei chen rulilong li christoph feichtenhofer jitendra malik shiry ginosar project page arxiv we trained a masked autoencoder with 3d gaussians as intermediate representations this allows the models to do zero shot figure ground segmentation image layering edge detection while performing same as supervised finetuning tasks scaling properties of diffusion models for perceptual tasks rahul ravishankar zeeshan patel jathushan rajasegaran jitendra malik computer vision and pattern recognition cvpr 2025 nbsp project page arxiv code we show how diffusion models benefit from scaling training and test time compute for perceptual tasks and unify tasks such as depth estimation optical flow and amodal segmentation under the framework of image to image translation humanoid locomotion as next token prediction ilija radosavovic bike zhang baifeng shi jathushan rajasegaran sarthak kamat trevor darrell koushil sreenath jitendra malik conference on neural information processing systems neurips 2024 nbsp project page arxiv real world humanoid control as a next token prediction problem akin to predicting the next word in language humans in 4d reconstructing and tracking humans with transformers shubham goel georgios pavlakos jathushan rajasegaran angjoo kanazawa jitendra malik international conference on computer vision iccv 2023 nbsp project page arxiv code demo a fully transformerized design for human mesh recovery achieves improved precision and remarkable robustness for 3d human reconstruction and tracking on the benefits of 3d pose and tracking for human action recognition jathushan rajasegaran georgios pavlakos angjoo kanazawa christoph feichtenhofer jitendra malik computer vision and pattern recognition cvpr 2023 nbsp project page paper arxiv code demo poster using 3d human reconstruction and tracking to recognize atomic actions in video tracking people by predicting 3d appearance location and pose jathushan rajasegaran georgios pavlakos angjoo kanazawa jitendra malik computer vision and pattern recognition cvpr 2022 oral presentation best paper finalist top 0 4 paper arxiv project page video results poster code performing monocular tracking of people by predicting their appearance pose and location and in 3d tracking people with 3d representations jathushan rajasegaran georgios pavlakos angjoo kanazawa jitendra malik neural information processing systems neurips 2021 paper arxiv project page video code poster performing monocular tracking of people by lifting them to 3d and then using 3d representations of their appearance pose and location itaml an incremental task agnostic meta learning approach jathushan rajasegaran salman khan munawar hayat fahad shahbaz khan computer vision and pattern recognition cvpr 2020 paper arxiv slides video code by learning generic representations from past tasks we can easily adapt to new tasks as well as remember old tasks random path selection for incremental learning jathushan rajasegaran munawar hayat salman khan fahad shahbaz khan ling shao neural information processing systems neurips 2019 paper arxiv poster code we increase the width of a resnet like model by adding extra skip connections when new tasks are introduced deepcaps going deeper with capsule networks jathushan rajasegaran vinoj jayasundara sandaru jayasekara hirunima jayasekara suranga seneviratne ranga rodrigo computer vision and pattern recognition cvpr 2019 nbsp oral presentation paper poster video code capsule networks are cool but they are shallow we can increase the depth by 3d convolutions and skip connections textcaps handwritten character recognition with very small datasets vinoj jayasundara sandaru jayasekara hirunima jayasekara jathushan rajasegaran suranga seneviratne ranga rodrigo winter conference on applications of computer vision wacv 2019 paper arxiv poster code capsule networks can capture actual variations that are present in human hand writing so we generate more data and retrain the capsule networks a multi modal neural embeddings approach for detecting mobile counterfeit apps a case study on google play store jathushan rajasegaran naveen karunanayake ashanie gunathillake suranga seneviratne guillaume jourjon international world wide web conference www 2019 paper arxiv poster we use content and style representations detect counterfeit apps in playstore website source from jon barron here