Meta tags:

Headings (most frequently used words):

-

Text of the page (most frequently used words):
and (32), the (25), for (24), yash (15), bhalgat (15), learning (12), method (12), with (12), 2024 (12), accepted (12), research (10), 2020 (10), nbsp (10), our (9), hosted (8), efficient (8), from (7), music (7), workshop (7), neurips (7), quantization (7), 2023 (7), 2025 (7), india (6), that (6), segmentation (6), cvpr (6), paper (6), also (5), deep (5), computer (5), vision (5), scene (5), feature (5), propose (5), lsq (5), andrew (5), zisserman (5), joão (5), henriques (5), contrastive (5), here (4), videos (4), all (4), university (4), show (4), object (4), data (4), using (4), 2019 (4), present (4), low (4), bit (4), aware (4), pruning (4), which (4), neural (4), 2022 (4), multi (4), geometry (4), code (4), lift (4), andrea (4), vedaldi (4), iro (4), laina (4), n2f2 (4), tracking (4), indian (3), this (3), among (3), rank (3), 2013 (3), bangalore (3), understanding (3), time (3), medical (3), supervision (3), teacher (3), student (3), training (3), are (3), approaches (3), model (3), results (3), nojun (3), kwak (3), jinwon (3), lee (3), learned (3), initialization (3), 2021 (3), network (3), fast (3), machine (3), fatih (3), porikli (3), structured (3), pose (3), epipolar (3), view (3), approach (3), instance (3), egocentric (3), work (3), github (3), website (2), youtube (2), channel (2), have (2), bachelors (2), iit (2), bombay (2), global (2), scholarship (2), selected (2), top (2), national (2), iitjee (2), exam (2), million (2), students (2), hall (2), fame (2), applications (2), shubham (2), jain (2), tcs (2), lab (2), google (2), invited (2), talks (2), scattering (2), seismic (2), based (2), performance (2), networks (2), image (2), cost (2), minimization (2), mixed (2), tri (2), thresholds (2), well (2), reduces (2), effective (2), methods (2), arxiv (2), qkd (2), end (2), sota (2), efficientnet (2), accuracy (2), tijmen (2), blankevoort (2), threshold (2), introduce (2), scheme (2), scale (2), parameters (2), driven (2), into (2), sylvester (2), has (2), free (2), especially (2), iclr (2), weight (2), loss (2), enables (2), decomposition (2), convolutions (2), its (2), estimations (2), during (2), iterative (2), gating (2), wacv (2), john (2), yang (2), hand (2), estimation (2), guided (2), transformer (2), predictions (2), light (2), touch (2), transformers (2), slow (2), fusion (2), spotlight (2), presentation (2), nested (2), fields (2), hierarchical (2), like (2), eccv (2), accv (2), scholar (2), available (2), senior (2), researcher (2), qualcomm (2), phd (2), oxford (2), funded (2), epsrc (2), aws (2), fellowship (2), aims (2), cdt (2), been (2), llm (2), news (2), want (2), llms (2), template, borrowed, office, hours, instagram, play, percussion, instrument, bachelor, classical, briefly, tried, learn, piano, harmonica, natural, beatboxer, sometimes, post, check, out, visharad, tabla, undergraduate, award, ura02, thesis, cargill, 2014, member, cohort, represent, seminar, minneapolis, usa, 2016, 300, compete, physics, chemistry, mathematics, olympiads, awarded, kvpy, govt, featured, international, olympiad, astronomy, astrophysics, advanced, 155, mains, spark, indabax, ghana, ghanaian, context, avataar, goel, pune, shirish, karande, val, iisc, venkatesh, babu, cvit, iiit, hyderabad, vineet, gandhi, makarand, tapaswi, peer, talk, peking, turing, class, committee, singapore, mike, shou, adobe, balaji, vasan, srinivasan, deepmind, prateek, slides, centric, virtual, video, 2018, use, wavelets, transforms, extract, sparse, sets, combined, simple, pca, selection, leads, promising, classification, affordable, computation, icassp, laurent, duval, jean, charlety, catseyes, categorizing, structures, tessellated, wavelet, budget, framework, setting, via, dense, segmentations, bounding, boxes, landmarks, imaging, meets, suyash, awate, meet, shah, annotation, suggestive, fully, convolutional, semi, supervised, classifiers, working, adaptive, konvens, amita, misra, jalal, mahmud, pritam, gundecha, zhe, liu, paradigm, unlabeled, exploitation, often, don, together, but, both, important, memory, footprint, combine, these, two, outperform, existing, chirag, patel, jangho, kim, knowledge, distillation, differentiable, layerwise, compression, ratios, alexnet, resnet, generates, trail, checkpoints, different, efficiency, operating, points, kambiz, azarian, general, asymmetric, trainable, offset, shows, mixnet, outperforming, markus, nagel, improving, through, learnable, offsets, better, initialize, cast, optimization, problem, restructured, known, equation, gradient, solutions, proposed, few, shot, fine, tuning, settings, practical, developing, countries, debasmit, das, solvers, neat, trick, enable, execution, convolution, operations, form, scaled, sum, pooling, components, structural, regularization, negligible, competitive, other, tensor, jamie, lin, yizhe, zhang, design, tiny, partial, layers, recursively, exploited, refining, previous, refinements, employ, criteria, decide, whether, exit, sharing, loop, allowing, per, sample, adaptation, predict, exploit, uncertainty, mechanism, simyung, chang, dynamic, refinement, incorporate, geometric, priors, models, can, implemented, test, implicitly, estimates, given, images, uses, downstream, invariant, retrieval, 150, lines, teaching, novel, scalable, achieving, significant, improvements, without, requiring, upper, bound, number, objects, encodes, properties, unified, field, outperforms, state, art, lerf, langsplat, open, vocabulary, tasks, complex, queries, while, enabling, faster, inference, leveraging, handle, rapid, motion, occlusions, improves, switches, reconstruction, amodal, long, vadim, tschernezki, full, list, publications, 3rd, position, micronet, competition, imagenet, track, leaderboard, preprint, got, promoted, started, dphil, intern, serving, chair, bmvc, hits, 900, stars, hashnerf, pytorch, organizing, 2nd, cvpr2024, icra, silvr, nefes, proud, mentor, moment, 3dv, reflecting, reality, gsloc, organized, successful, 1st, vla, semantic, correspondence, jamais, linkedin, email, feel, you, discuss, ideas, around, startups, collaborate, setup, call, electrical, engineering, minor, masters, science, michigan, education, parallelly, consultant, startup, company, before, was, fortunate, spend, ibm, almaden, ifpen, paris, voxel51, content, moderation, device, ltp, structconv, language, dir, net, interests, final, year, candidate, advised, visual, group, vgg, भळगट, blog,

Text of the page (random words):
yash bhalgat news research invited talks hall of fame music blog यश भळगट yash bhalgat final year phd candidate in the visual geometry group vgg oxford co advised by andrew zisserman andrea vedaldi joão henriques and iro laina funded by the epsrc aws fellowship with aims cdt research interests 3d computer vision contrastive lift light touch dir net vision language n2f2 3d llms efficient machine learning lsq qkd structconv ltp parallelly i also work as an ai consultant for an on device ai startup and a llm content moderation company before i was a senior researcher at qualcomm ai research i have also been fortunate to spend time at voxel51 ibm research bangalore and almaden lab ifpen paris tcs research education masters in computer science from the university of michigan bachelors in electrical engineering cs minor from iit bombay feel free to setup a call if you want to discuss ideas around startups ai cv ml llms or want to collaborate email nbsp nbsp cv nbsp nbsp scholar nbsp nbsp github nbsp nbsp linkedin nbsp nbsp x news 09 2025 jamais vu our paper on semantic correspondence learning accepted to neurips 2025 06 2025 organized a successful 1st workshop on 3d llm vla at cvpr 2025 01 2025 gsloc accepted to iclr 2025 11 2024 proud mentor moment reflecting reality accepted to 3dv 2025 09 2024 our 3d aware egocentric tracking paper accepted to accv 2024 07 2024 n2f2 accepted to eccv 2024 02 2024 nefes accepted to cvpr 2024 01 2024 silvr accepted to icra 2024 01 2024 we are organizing the 2nd workshop on learning 3d with multi view supervision at cvpr2024 09 2023 contrastive lift paper accepted to neurips 2023 as a spotlight presentation code available github 03 2023 epipolar guided transformers paper accepted to cvpr 2023 01 2023 hashnerf pytorch hits 900 stars on github 06 2022 serving as a website chair for bmvc 2022 10 2021 3d hand pose estimation work with my intern john yang has been accepted to wacv 2022 10 2021 started my dphil phd at university of oxford in aims cdt funded by epsrc aws fellowship 11 2020 i got promoted to senior machine learning researcher at qualcomm ai research 09 2020 structured convolutions paper accepted to neurips 2020 03 2020 lsq paper accepted to the efficient deep learning in computer vision workshop at cvpr 2020 02 2020 preprint for our work on learned threshold pruning available 11 2019 3rd position in neurips 2019 micronet competition imagenet track leaderboard code here and here selected publications full list google scholar 3d aware instance segmentation and tracking in egocentric videos yash bhalgat vadim tschernezki iro laina joão henriques andrea vedaldi andrew zisserman accv 2024 we propose a 3d aware method for object tracking in long egocentric videos leveraging scene geometry to handle rapid motion and occlusions our approach improves tracking accuracy reduces id switches by up to 80 and enables applications like 3d object reconstruction and amodal segmentation n2f2 hierarchical scene understanding with nested neural feature fields yash bhalgat iro laina joão henriques andrew zisserman andrea vedaldi eccv 2024 we present nested neural feature fields n2f2 a hierarchical approach to 3d scene understanding that encodes multi scale properties in a unified feature field our method outperforms state of the art approaches like lerf and langsplat on open vocabulary 3d tasks especially for complex queries while enabling faster inference contrastive lift 3d object instance segmentation by slow fast contrastive fusion code yash bhalgat iro laina joão henriques andrew zisserman andrea vedaldi neurips 2023 spotlight presentation we present a novel slow fast contrastive fusion method to lift 2d predictions to 3d for scalable instance segmentation achieving significant improvements without requiring an upper bound on the number of objects in the scene a light touch approach to teaching transformers multi view geometry yash bhalgat joão henriques andrew zisserman cvpr 2023 an epipolar guided training method to incorporate multi view geometric priors into transformer models which can be implemented in 150 lines of code during test time the transformer implicitly estimates the epipolar geometry given 2 images and uses it for downstream predictions e g for pose invariant retrieval dynamic iterative refinement for efficient 3d hand pose estimation john yang yash bhalgat simyung chang fatih porikli nojun kwak wacv 2022 we propose a tiny deep network of which partial layers are recursively exploited for refining its previous estimations during its iterative refinements we employ learned gating criteria to decide whether to exit from the weight sharing loop allowing per sample adaptation in our model we also predict and exploit uncertainty estimations in the gating mechanism structured convolutions for efficient neural network design yash bhalgat yizhe zhang jamie lin fatih porikli neurips 2020 we introduce a neat trick to enable the execution of convolution operations in the form of efficient scaled sum pooling components we present a structural regularization loss that enables this decomposition with negligible performance loss our method is competitive with other tensor decomposition and structured pruning methods data driven weight initialization with sylvester solvers debasmit das yash bhalgat fatih porikli practical machine learning for developing countries workshop iclr 2021 we propose a data driven scheme to initialize the parameters of a neural network the initialization is cast as an optimization problem which is restructured into the well known sylvester equation that has fast and efficient gradient free solutions we show that our proposed method is especially effective in few shot and fine tuning settings lsq improving low bit quantization through learnable offsets and better initialization yash bhalgat jinwon lee markus nagel tijmen blankevoort nojun kwak efficient deep learning in computer vision workshop cvpr 2020 we introduce a general asymmetric quantization scheme with trainable scale and offset parameters lsq shows sota results for efficientnet and mixnet outperforming lsq for low bit quantization learned threshold pruning kambiz azarian yash bhalgat jinwon lee tijmen blankevoort arxiv 2020 we propose an end to end differentiable method for learning layerwise pruning thresholds which results in sota model compression ratios with alexnet resnet and efficientnet our method also generates a trail of checkpoints with different accuracy efficiency operating points qkd quantization aware knowledge distillation for low bit quantization yash bhalgat jangho kim jinwon lee chirag patel nojun kwak arxiv 2020 low bit quantization and kd often don t go well together but both are important approaches to reduces a model s memory footprint we propose an effective method to combine these two methods and show results that outperform all existing quantization kd approaches teacher student learning paradigm for tri training an efficient method for unlabeled data exploitation yash bhalgat zhe liu pritam gundecha jalal mahmud amita misra konvens 2019 teacher student tri training is a method for semi supervised learning using 3 classifiers working using adaptive teacher and student thresholds annotation cost minimization for medical image segmentation using suggestive mixed supervision fully convolutional networks yash bhalgat meet shah suyash awate medical imaging meets neurips workshop 2019 for medical image segmentation we present a budget based cost minimization framework in a mixed supervision setting via dense segmentations bounding boxes and landmarks catseyes categorizing seismic structures with tessellated scattering wavelet networks yash bhalgat jean charlety laurent duval icassp 2018 we use scattering wavelets transforms to extract sparse feature sets from seismic data we show that using this method combined with simple pca based feature selection leads to promising classification performance in affordable computation time invited talks from video to virtual object centric 3d scene understanding from videos slides google deepmind india hosted by prateek jain adobe research bangalore hosted by balaji vasan srinivasan show lab national university of singapore hosted by mike shou cs peer talk peking university hosted by turing class research committee cvit iiit hyderabad hosted by makarand tapaswi and vineet gandhi val iisc bangalore hosted by r venkatesh babu tcs research pune hosted by shirish karande avataar ai hosted by shubham goel and shubham jain deep learning for computer vision and applications to the ghanaian context youtube spark deep learning workshop at indabax ghana hall of fame all india rank 12 in iitjee mains 2013 exam among 1 5 million students all india rank 155 in iitjee advanced 2013 exam among 0 2 million students featured in national top 30 for the international olympiad on astronomy and astrophysics 2013 all india rank 60 and awarded the kvpy scholarship by govt of india among top 300 in india to compete in the physics chemistry and mathematics olympiads cargill global scholarship 2014 15 and selected in the 10 member indian cohort to represent at the global seminar in minneapolis usa in 2016 undergraduate research award ura02 for bachelors thesis at iit bombay music i play the tabla an indian percussion instrument and have a visharad bachelor of music in indian classical music i ve also briefly tried to learn the piano and harmonica i am a natural beatboxer check this out and i sometimes post music videos here instagram music channel youtube music channel office hours website template borrowed from here