Meta tags:

Headings (most frequently used words):

the, 3d, and, in, to, of, we, for, with, is, physical, on, temporal, by, generation, real, based, world, propose, transformer, assets, from, tracking, model, knowledge, large, physx, that, this, generative, categories, performance, our, features, ziang, cao, tctrack, single, liang, abstract, existing, quality, an, aware, contexts, similarity, aerial, are, can, framework, high, geometry, it, diffusion, paper, accepted, one, difftf, two, simulation, pan, ziwei, liu, introduce, efficiently, which, rich, extensive, experiments, vision, generalized, refinement, feature, as, anything, visual, be, first, specifically, learning, diversity, object, demonstrate, physics, grounded, textbf, end, efficient, texture, triplane, multi, maps, map, hift, ready, chen, most, methods, thereby, their, ai, more, objects, across, feed, forward, vocabulary, diverse, adopt, enhance, information, frames, adaptive, convolution, target, cvpr, 2025, 2022, 2026, at, tpami, 2021, image, fangzhou, hong, modeling, used, properties, embodied, new, representation, datasets, dataset, over, validate, asset, while, despite, models, devise, into, yet, task, only, or, few, challenges, handle, specialized, decoder, reconstruction, effectiveness, state, art, semantics, speed, attention, changhong, fu, conditions, among, fully, level, extraction, spatial, using, evaluation, benchmarks, tests, egocentric, li, its, hierarchical, layers, feb, 2024, zhaoxi, directly, however, overlook, articulation, limiting, bridge, gap, wild, sim, explicit, vlm, tokens, introducing, fine, improving, addition, mobility, than, strong, generalization, furthermore, applications, especially, neglecting, consequently, often, important, like, address, paradigm, annotated, present, scale, human, annotation, physxgen, space, architecture, between, all, code, research, tong, wu, generating, challenging, efforts, meanwhile, focus, category, complexity, efficiency, encoder, stronger, version, boils, down, view, loss, additionally, triplanes, shapenet, omniobject3d, convincingly, proposed, automatic, three, appearances, novel, handling, improve, accuracy, cross, plane, complex, https, github, ziyuan, huang, shiwei, zhang, has, trackers, consecutive, far, being, utilized, exploit, levels, temporally, achieved, dynamically, calibrating, weights, accurate, further, also, online, action, yiming, but, not, communities, work, comprehensive, effective, four, convolutional, have, selected, awards, scholarships, papers, neurips, spotlight, iclr, publications, education, experiences, apr, china3dv, awarded, may, aug, shifting, static, representations, toward, articulated, interaction, key, utility, given, produces, attributes, along, tokenizes, reduces, number, 193x, enabling, within, standard, ken, budgets, without, any, special, during, tuning, significantly, overcome, limited, construct, expands, prior, phys, ical, 2x, includes, 2k, common, annotations, images, delivers, robust, mujoco, style, environment, di, rectly, contact, robotic, policy, believe, substantially, empower, broad, range, downstream, moving, virtual, primarily, emphasizes, geometries, textures, rapid, development, synthesized, hampering, application, domains, initial, attempt, challenge, critical, physxnet, systematically, five, foundational, dimensions, absolute, material, affordance, kinematics, function, description, particular, scalable, loop, pipeline, language, enables, creation, raw, injecting, pre, trained, structural, employs, dual, branch, explicitly, latent, correlations, structures, producing, plausible, predictions, preserving, native, superior, promising, capability, data, will, released, facilitate, future, automatically, poses, fundamental, computer, optimization, approaches, struggle, produce, generalizability, therefore, these, improved, guarantee, aggregate, building, upon, parts, utilize, tune, avoiding, negative, influence, caused, errors, synthesis, eliminating, mismatch, stages, enhanced, process, introduced, filter, out, artifacts, refine, resulting, intricate, reasonable, details, modules, creating, highly, desirable, works, synthesizing, massive, notably, there, major, need, expressive, via, aspects, considering, robustness, revised, fitting, drastic, variations, regard, combination, extract, shared, learns, relations, different, planes, aggregates, encoded, 200, achieves, project, page, ziangcao0312, io, difftf_pages, towards, made, significant, improvements, past, decades, merely, aim, ideal, overlooking, detection, integrate, template, where, those, problems, encode, decode, curriculum, strategy, measure, exhaustive, wellknown, superiority, verify, readily, prediction, andrew, benjamin, luoyao, hang, zhao, feng, denotes, equal, contribution, interested, anticipating, early, possible, location, person, manipulation, workspace, fields, robot, collaboration, received, enough, stimulate, modal, million, rgbd, imu, streams, provide, metrics, 2d, labels, semi, design, baseline, recurrent, neural, networks, rnns, conduct, various, ablation, studies, results, worthy, study, researchers, robotics, incorporated, according, previous, effectively, encodes, memory, way, before, decoded, adjustment, shows, impressive, uav, show, 27, fps, nvidia, jetson, agx, xavier, junjie, ye, bowen, iccv, siamese, execute, classification, regression, they, either, employ, last, layer, degrades, localization, scenarios, separately, use, multiple, decision, making, intractable, computations, mobile, platforms, thus, generated, fed, achieve, interactive, fusion, shallow, cues, deep, global, contextual, raised, facilitating, search, learn, interdependencies, discovering, tailored, discriminability, evaluations, proven, platform, strongly, validated, practicability, time, available, com, vision4robotics, news, invited, talk, rising, star, award, top, oct, am, google, phd, fellowship, machine, perception, sep, 3dtopia, xl, cvpr2025, highlight, jan, 2023, mar, egopat3d, cvpr2022, ra, l2022, icra2022, iccv2021, july, siamapn, darklighter, iros2021, tgrs, icra2021, about, me, cv,

Text of the page (most frequently used words):
the (81), and (55), for (24), with (21), #physical (18), #temporal (16), generation (15), real (14), based (13), world (13), accepted (13), propose (12), transformer (12), paper (12), one (12), tracking (11), from (11), assets (11), physx (11), ziang (10), cao (10), knowledge (10), large (10), model (10), this (9), our (9), features (9), tctrack (9), performance (9), that (9), generative (9), categories (9), 2022 (8), existing (8), similarity (8), single (8), aerial (8), are (8), abstract (8), contexts (8), liang (8), vision (8), quality (8), aware (8), feb (7), 2021 (7), 2025 (7), can (7), framework (7), two (7), high (7), ziwei (7), liu (7), learning (7), diffusion (7), geometry (7), difftf (7), 2026 (6), which (6), feature (6), efficiently (6), refinement (6), pan (6), rich (6), introduce (6), experiments (6), extensive (6), generalized (6), simulation (6), anything (6), 2023 (5), top (5), object (5), maps (5), map (5), efficient (5), hift (5), multi (5), end (5), visual (5), textbf (5), specifically (5), first (5), demonstrate (5), diversity (5), texture (5), triplane (5), grounded (5), physics (5), shanghai (4), scholarship (4), research (4), most (4), methods (4), target (4), information (4), thereby (4), frames (4), adaptive (4), convolution (4), enhance (4), cvpr (4), more (4), their (4), chen (4), adopt (4), tpami (4), diverse (4), feed (4), forward (4), objects (4), vocabulary (4), across (4), computer (4), ready (4), present (3), nanyang (3), technological (3), university (3), china (3), national (3), champion (3), design (3), challenge (3), star (3), student (3), omniobject3d (3), phd (3), china3dv (3), hierarchical (3), level (3), layers (3), into (3), spatial (3), semantics (3), deep (3), only (3), among (3), benchmarks (3), effectiveness (3), tests (3), its (3), speed (3), github (3), changhong (3), fully (3), extraction (3), using (3), evaluation (3), over (3), egocentric (3), yet (3), attention (3), task (3), dataset (3), validate (3), new (3), few (3), state (3), art (3), conditions (3), while (3), handle (3), used (3), despite (3), challenges (3), representation (3), specialized (3), devise (3), decoder (3), fangzhou (3), hong (3), 2024 (3), reconstruction (3), modeling (3), models (3), properties (3), embodied (3), asset (3), datasets (3), image (3), papers (3), jun (2), sep (2), 2018 (2), second (2), undergraduate (2), 2019 (2), iccv (2), google (2), fellowship (2), rising (2), award (2), however (2), convolutional (2), accuracy (2), complex (2), introducing (2), work (2), effective (2), consequently (2), not (2), but (2), also (2), architecture (2), space (2), strong (2), comprehensive (2), four (2), have (2), code (2), https (2), yiming (2), consecutive (2), far (2), being (2), utilized (2), trackers (2), exploit (2), levels (2), online (2), temporally (2), proposed (2), achieved (2), dynamically (2), calibrating (2), weights (2), accurate (2), ziyuan (2), huang (2), shiwei (2), zhang (2), action (2), important (2), like (2), human (2), has (2), communities (2), challenging (2), than (2), automatic (2), annotation (2), meanwhile (2), further (2), robotics (2), paradigm (2), neglecting (2), stronger (2), version (2), boils (2), down (2), improve (2), additionally (2), directly (2), applications (2), efforts (2), focus (2), category (2), three (2), complexity (2), appearances (2), novel (2), handling (2), efficiency (2), all (2), cross (2), plane (2), addition (2), encoder (2), triplanes (2), shapenet (2), convincingly (2), tong (2), iclr (2), generating (2), scale (2), often (2), limiting (2), address (2), view (2), loss (2), fine (2), improving (2), between (2), especially (2), overlook (2), bridge (2), gap (2), annotated (2), furthermore (2), physxgen (2), generalization (2), data (2), zhaoxi (2), neurips (2), spotlight (2), articulation (2), wild (2), sim (2), explicit (2), vlm (2), tokens (2), mobility (2), selected (2), may (2), aug (2), awarded (2), apr (2), interests (2), singapore, artificial, intelligence, laboratory, vehicle, engineering, 2017, tonji, education, experiences, prize, chinese, advanced, mathematics, competition, honda, eco, mileage, 2020, baosteel, tongji, excellent, solution, awards, scholarships, siamese, execute, classification, regression, they, either, employ, last, layer, degrades, localization, scenarios, separately, use, multiple, decision, making, intractable, computations, mobile, platforms, thus, generated, fed, achieve, interactive, fusion, shallow, cues, global, contextual, raised, facilitating, search, learn, interdependencies, discovering, tailored, discriminability, evaluations, proven, platform, strongly, validated, practicability, time, available, com, vision4robotics, junjie, bowen, incorporated, according, previous, effectively, encodes, memory, way, before, decoded, adjustment, shows, impressive, uav, show, fps, nvidia, jetson, agx, xavier, interested, anticipating, early, possible, location, person, manipulation, workspace, fields, robot, collaboration, received, enough, stimulate, modal, million, rgbd, imu, streams, provide, metrics, labels, semi, baseline, recurrent, neural, networks, rnns, conduct, various, ablation, studies, results, worthy, study, researchers, andrew, benjamin, luoyao, hang, zhao, feng, denotes, equal, contribution, prediction, made, significant, improvements, past, decades, merely, aim, ideal, overlooking, detection, integrate, template, where, those, problems, encode, decode, curriculum, strategy, measure, exhaustive, wellknown, superiority, verify, readily, towards, creating, highly, desirable, works, synthesizing, massive, notably, there, major, need, expressive, via, aspects, considering, robustness, revised, fitting, drastic, variations, regard, combination, extract, shared, learns, relations, different, planes, aggregates, encoded, 200, achieves, project, page, ziangcao0312, difftf_pages, automatically, poses, fundamental, optimization, approaches, struggle, produce, generalizability, therefore, these, improved, guarantee, aggregate, building, upon, parts, utilize, tune, avoiding, negative, influence, caused, errors, synthesis, eliminating, mismatch, stages, enhanced, process, introduced, filter, out, artifacts, refine, resulting, intricate, reasonable, details, modules, moving, virtual, primarily, emphasizes, geometries, textures, rapid, development, synthesized, hampering, application, domains, initial, attempt, critical, physxnet, systematically, five, foundational, dimensions, absolute, material, affordance, kinematics, function, description, particular, scalable, loop, pipeline, language, enables, creation, raw, injecting, pre, trained, structural, employs, dual, branch, explicitly, latent, correlations, structures, producing, plausible, predictions, preserving, native, superior, promising, capability, will, released, facilitate, future, shifting, static, representations, toward, articulated, interaction, key, utility, given, produces, attributes, along, tokenizes, reduces, number, 193x, enabling, within, standard, ken, budgets, without, any, special, during, tuning, significantly, overcome, limited, construct, expands, prior, phys, ical, includes, common, annotations, images, delivers, robust, mujoco, style, environment, rectly, contact, robotic, policy, believe, substantially, empower, broad, range, downstream, publications, icra2021, tgrs, july, iros2021, darklighter, siamapn, iccv2021, icra2022, l2022, mar, cvpr2022, egopat3d, jan, cvpr2025, highlight, 3dtopia, oct, machine, perception, invited, talk, news, currently, year, college, science, unviersity, supervised, his, lie, prof, mmlab, ntu, about, twitter, orcid, scholar, email, social, ziangcao, homepage,

Text of the page (random words):
ziangcao homepage ziang cao ph d student nanyang technological university cv social email scholar orcid github twitter interests deep learning robotics computer vision about me ziang cao is currently a second year phd student in the college of computer and data science in nanyang technological unviersity with mmlab ntu supervised by prof ziwei liu his research interests lie on the computer vision deep learning and 3d generation news apr 2026 invited talk at china3dv and awarded rising star award apr 2026 physx anything is selected as one of the top 5 papers at china3dv feb 2026 one paper physx anything is accepted by cvpr 2026 oct 2025 i am awarded google phd fellowship 2025 machine perception sep 2025 one paper physx 3d is accepted by neurips 2025 as spotlight feb 2025 one paper 3dtopia xl is accepted by cvpr2025 as highlight may 2024 one paper difftf is accepted by tpami jan 2024 one paper difftf is accepted by iclr aug 2023 one paper tctrack is accepted by tpami mar 2022 two papers tctrack and egopat3d are accepted by cvpr2022 feb 2022 one paper is accepted by ra l2022 feb 2022 one paper is accepted by icra2022 aug 2021 one paper hift is accepted by iccv2021 july 2021 two papers siamapn and darklighter are accepted by iros2021 may 2021 one paper is accepted by tgrs feb 2021 one paper is accepted by icra2021 selected publications physx anything simulation ready physical 3d assets from single image ziang cao fangzhou hong zhaoxi chen liang pan ziwei liu cvpr 2026 abstract 3d modeling is shifting from static visual representations toward physical articulated assets that can be directly used in simulation and interaction however most existing 3d generation methods overlook key physical and articulation properties thereby limiting their utility in embodied ai to bridge this gap we introduce physx anything the first simulation ready physical 3d generative framework that given a single in the wild image produces high quality sim ready 3d assets with explicit geometry articulation and physical attributes specifically we propose the first vlm based physical 3d generative model along with a new 3d representation that efficiently tokenizes geometry it reduces the number of tokens by 193x enabling explicit geometry learning within standard vlm to ken budgets without introducing any special tokens during fine tuning and significantly improving generative quality in addition to overcome the limited diversity of existing physical 3d datasets we construct a new dataset physx mobility which expands the object categories in prior phys ical 3d datasets by over 2x and includes more than 2k common real world objects with rich physical annotations extensive experiments on physx mobility and in the wild images demonstrate that physx anything delivers strong generative performance and robust generalization furthermore simulation based experiments in a mujoco style environment validate that our sim ready assets can be di rectly used for contact rich robotic policy learning we believe physx anything can substantially empower a broad range of downstream applications especially in embodied ai and physics based simulation physx 3d physical grounded 3d asset generation ziang cao zhaoxi chen liang pan ziwei liu neurips 2025 spotlight abstract 3d modeling is moving from virtual to physical existing 3d generation primarily emphasizes geometries and textures while neglecting physical grounded modeling consequently despite the rapid development of 3d generative models the synthesized 3d assets often overlook rich and important physical properties hampering their real world application in physical domains like simulation and embodied ai as an initial attempt to address this challenge we propose textbf physx 3d an end to end paradigm for physical grounded 3d asset generation 1 to bridge the critical gap in physics annotated 3d datasets we present physxnet the first physics grounded 3d dataset systematically annotated across five foundational dimensions absolute scale material affordance kinematics and function description in particular we devise a scalable human in the loop annotation pipeline based on vision language models which enables efficient creation of physics first assets from raw 3d assets 2 furthermore we propose textbf physxgen a feed forward framework for physics grounded image to 3d asset generation injecting physical knowledge into the pre trained 3d structural space specifically physxgen employs a dual branch architecture to explicitly model the latent correlations between 3d structures and physical properties thereby producing 3d assets with plausible physical predictions while preserving the native geometry quality extensive experiments validate the superior performance and promising generalization capability of our framework all the code data and models will be released to facilitate future research in generative physical ai difftf 3d aware diffusion transformer for large vocabulary 3d generation ziang cao fangzhou hong tong wu liang pan ziwei liu tpami abstract generating diverse and high quality 3d assets automatically poses a fundamental yet challenging task in 3d computer vision despite extensive efforts in 3d generation existing optimization based approaches struggle to produce large scale 3d assets efficiently meanwhile feed forward methods often focus on generating only a single category or a few categories limiting their generalizability therefore we introduce a diffusion based feed forward framework to address these challenges with a single model to handle the large diversity and complexity in geometry and texture across categories efficiently we 1 adopt improved triplane to guarantee efficiency 2 introduce the 3d aware transformer to aggregate the generalized 3d knowledge with specialized 3d features and 3 devise the 3d aware encoder decoder to enhance the generalized 3d knowledge building upon our 3d aware diffusion model with transformer difftf we propose a stronger version for 3d generation i e difftf it boils down to two parts multi view reconstruction loss and triplane refinement specifically we utilize multi view reconstruction loss to fine tune the diffusion model and triplane decoder thereby avoiding the negative influence caused by reconstruction errors and improving texture synthesis by eliminating the mismatch between the two stages the generative performance is enhanced especially in texture additionally a 3d aware refinement process is introduced to filter out artifacts and refine triplanes resulting in the generation of more intricate and reasonable details extensive experiments on shapenet and omniobject3d convincingly demonstrate the effectiveness of our proposed modules and the state of the art 3d object generation performance with large diversity rich semantics and high quality large vocabulary 3d diffusion model with transformer ziang cao fangzhou hong tong wu liang pan ziwei liu iclr 2024 abstract creating diverse and high quality 3d assets with an automatic generative model is highly desirable despite extensive efforts on 3d generation most existing works focus on the generation of a single category or a few categories in this paper we introduce a diffusion based feed forward framework for synthesizing massive categories of real world 3d objects with a single generative model notably there are three major challenges for this large vocabulary 3d generation a the need for expressive yet efficient 3d representation b large diversity in geometry and texture across categories c complexity in the appearances of real world objects to this end we propose a novel triplane based 3d aware diffusion model with transformer difftf for handling challenges via three aspects 1 considering efficiency and robustness we adopt a revised triplane representation and improve the fitting speed and accuracy 2 to handle the drastic variations in geometry and texture we regard the features of all 3d objects as a combination of generalized 3d knowledge and specialized 3d features to extract generalized 3d knowledge from diverse categories we propose a novel 3d aware transformer with shared cross plane attention it learns the cross plane relations across different planes and aggregates the generalized 3d knowledge with specialized 3d features 3 in addition we devise the 3d aware encoder decoder to enhance the generalized 3d knowledge in the encoded triplanes for handling categories with complex appearances extensive experiments on shapenet and omniobject3d over 200 diverse real world categories convincingly demonstrate that a single difftf model achieves state of the art large vocabulary 3d object generation performance with large diversity rich semantics and high quality our project page https ziangcao0312 github io difftf_pages towards real world visual tracking with temporal contexts ziang cao ziyuan huang liang pan shiwei zhang ziwei liu changhong fu tpami abstract visual tracking has made significant improvements in the past few decades most existing state of the art trackers 1 merely aim for performance in ideal conditions while overlooking the real world conditions 2 adopt the tracking by detection paradigm neglecting rich temporal contexts 3 only integrate the temporal information into the template where temporal contexts among consecutive frames are far from being fully utilized to handle those problems we propose a two level framework tctrack that can exploit temporal contexts efficiently based on it we propose a stronger version for real world visual tracking i e tctrack it boils down to two levels features and similarity maps specifically for feature extraction we propose an attention based temporally adaptive convolution to enhance the spatial features using temporal information which is achieved by dynamically calibrating the convolution weights for similarity map refinement we introduce an adaptive temporal transformer to encode the temporal knowledge efficiently and decode it for the accurate refinement of the similarity map to further improve the performance we additionally introduce a curriculum learning strategy also we adopt online evaluation to measure performance in real world conditions exhaustive experiments on 8 wellknown benchmarks demonstrate the superiority of tctrack real world tests directly verify that tctrack can be readily used in real world applications egocentric prediction of action target in 3d yiming li ziang cao andrew liang benjamin liang luoyao chen hang zhao chen feng cvpr 2022 denotes equal contribution abstract we are interested in anticipating as early as possible the target location of a person s object manipulation action in a 3d workspace from egocentric vision it is important in fields like human robot collaboration but has not yet received enough attention from vision and learning communities to stimulate more research on this challenging egocentric vision task we propose a large multi modal dataset of more than 1 million frames of rgbd and imu streams and provide evaluation metrics based on our high quality 2d and 3d labels from semi automatic annotation meanwhile we design baseline methods using recurrent neural networks rnns and conduct various ablation studies to validate their effectiveness our results demonstrate that this new task is worthy of further study by researchers in robotics vision and learning communities tctrack temporal contexts for aerial tracking ziang cao ziyuan huang liang pan shiwei zhang ziwei liu changhong fu cvpr 2022 abstract temporal contexts among consecutive frames are far from being fully utilized in existing visual trackers in this work we present tctrack a comprehensive framework to fully exploit temporal contexts for aerial tracking the temporal contexts are incorporated at textbf two levels the extraction of textbf features and the refinement of textbf similarity maps specifically for feature extraction an online temporally adaptive convolution is proposed to enhance the spatial features using temporal information which is achieved by dynamically calibrating the convolution weights according to the previous frames for similarity map refinement we propose an adaptive temporal transformer which first effectively encodes temporal knowledge in a memory efficient way before the temporal knowledge is decoded for accurate adjustment of the similarity map tctrack is effective and efficient evaluation on four aerial tracking benchmarks shows its impressive performance real world uav tests show its high speed of over 27 fps on nvidia jetson agx xavier hift hierarchical feature transformer for aerial tracking ziang cao changhong fu junjie ye bowen li yiming li iccv 2021 abstract most existing siamese based tracking methods execute the classification and regression of the target object based on the similarity maps however they either employ a single map from the last convolutional layer which degrades the localization accuracy in complex scenarios or separately use multiple maps for decision making introducing intractable computations for aerial mobile platforms thus in this work we propose an efficient and effective hierarchical feature transformer hift for aerial tracking hierarchical similarity maps generated by multi level convolutional layers are fed into the feature transformer to achieve the interactive fusion of spatial shallow layers and semantics cues deep layers consequently not only the global contextual information can be raised facilitating the target search but also our end to end architecture with the transformer can efficiently learn the interdependencies among multi level features thereby discovering a tracking tailored feature space with strong discriminability comprehensive evaluations on four aerial benchmarks have proven the effectiveness of hift real world tests on the aerial platform have strongly validated its practicability with a real time speed our code is available at https github com vision4robotics hift awards scholarships 2026 china3dv rising star award 2025 google phd fellowship 2023 champion solution in omniobject3d challenge iccv 2023 champion design 2022 excellent undergraduate student of shanghai 2021 research star of tongji 2021 baosteel scholarship top 0 5 2020 national scholarship top 1 2019 honda china eco mileage challenge champion design 2019 national scholarship top 1 2018 second prize of chinese undergraduate advanced mathematics competition 2018 national scholarship top 1 education experiences sep 2017 jun 2022 tonji university vehicle engineering shanghai china jun 2022 feb 2023 shanghai artificial intelligence laboratory shanghai china feb 2023 present nanyang technological university singapore