Meta tags:

Headings (most frequently used words):

-

Text of the page (most frequently used words):
and (56), the (30), autoflow (24), training (24), flow (20), optical (19), for (18), supervised (13), self (12), pwc (10), sun (9), deqing (9), herrmann (9), charles (9), code (9), raft (9), better (8), set (8), dataset (8), pre (8), trained (8), models (7), results (7), net (7), newly (7), kitti (7), learning (6), michael (6), freeman (6), william (6), cvpr (6), with (6), sintel (6), our (6), examples (5), datasets (5), target (5), than (5), between (5), disentangling (5), architecture (5), ground (5), truth (5), back (4), part (4), performance (4), accuracy (4), available (4), published (4), 2015 (4), davis (4), top (3), inproceedings (3), title (3), author (3), vlasic (3), daniel (3), jampani (3), varun (3), krainin (3), chang (3), huiwen (3), zabih (3), ramin (3), liu (3), booktitle (3), year (3), 2021 (3), bibtex (3), link (3), cvf (3), arxiv (3), papers (3), they (3), are (3), data (3), where (3), abstract (3), left (3), learns (3), paper (3), google (3), research (3), reda (3), fitsum (3), rubinstein (3), fleet (3), david (3), eccv (3), 2022 (3), details (3), irr (3), more (3), all (3), methods (3), huang (3), hsin (3), ping (3), hur (3), junhwa (3), erika (3), sargent (3), kyle (3), stone (3), austin (3), yang (3), ming (3), hsuan (3), 2023 (3), search (3), metric (3), real (3), world (3), performs (3), static (2), samples (2), synthetic (2), but (2), generate (2), new (2), render (2), that (2), model (2), show (2), achieves (2), state (2), art (2), flyingchairs (2), right (2), augmentation (2), when (2), now (2), recent (2), explore (2), these (2), techniques (2), gains (2), large (2), improvements (2), original (2), accurate (2), time (2), writing (2), hole (2), cart (2), person (2), labels (2), domain (2), without (2), using (2), sun2021autoflow, 96g, total, license, 000, oral, presentation, play, critical, role, cnn, painstaking, hard, adapt, applications, automate, process, present, simple, effective, method, optimizes, takes, layered, approach, motion, shape, appearance, each, layer, controlled, learnable, hyperparameters, experimental, both, typical, pipeline, trains, then, evaluates, which, optimized, ona, four, achieve, lower, errors, 872, gap, becomes, small, enough, number, final, pipelines, sun2022disentangling, project, lead, equal, technical, contribution, how, important, like, generalize, questions, rather, develop, revisit, three, prominent, common, modern, observe, significant, demonstrating, importance, generality, surprisingly, versus, benchmarks, outperform, flow1d, while, being, faster, during, inference, score, demonstrate, benefits, separating, contributions, analyzing, originally, blue, red, shows, improved, recovers, front, visual, comparison, sequence, huang2023self, recently, has, shown, promising, requires, compute, its, observing, strong, correlation, losses, introduce, handle, videos, loss, par, further, semi, setting, obtain, competitive, against, through, supervision, comparable, requiring, not, report, keypoint, propagation,

Text of the page (random words):
autoflow autoflow learning a better training set for optical flow cvpr 2021 disentangling architecture and training for optical flow eccv 2022 self supervised autoflow cvpr 2023 self supervised autoflow hsin ping huang charles herrmann junhwa hur erika lu kyle sargent austin stone ming hsuan yang deqing sun google research paper code self supervised autoflow learns to generate an optical flow training set through self supervision on the target domain it performs comparable to supervised autoflow on sintel and kitti without requiring ground truth gt and learns a better dataset for real world davis where gt is not available we report optical flow accuracy on sintel and kitti and keypoint propagation accuracy on davis abstract recently autoflow has shown promising results on learning a training set for optical flow but requires ground truth labels in the target domain to compute its search metric observing a strong correlation between the ground truth search metric and self supervised losses we introduce self supervised autoflow to handle real world videos without ground truth labels using self supervised loss as the search metric our self supervised autoflow performs on par with autoflow on sintel and kitti where ground truth is available and performs better on the real world davis dataset we further explore using self supervised autoflow in the semi supervised setting and obtain competitive results against the state of the art papers self supervised autoflow hsin ping huang charles herrmann junhwa hur erika lu kyle sargent austin stone ming hsuan yang and deqing sun cvpr 2023 arxiv cvf code link to code for self supervised autoflow bibtex inproceedings huang2023self title self supervised autoflow author huang hsin ping and herrmann charles and hur junhwa and lu erika and sargent kyle and stone austin and yang ming hsuan and sun deqing booktitle cvpr year 2023 back to top disentangling architecture and training for optical flow deqing sun charles herrmann fitsum reda michael rubinstein david fleet william t freeman google research paper code left large improvements with newly trained pwc net irr pwc and raft left originally published results in blue results of our newly trained models in red the newly trained raft is more accurate than all published methods on kitti 2015 at the time of writing right visual comparison on a davis sequence between the original 43 and our newly trained pwc net and raft shows improved flow details e g the hole between the cart and the person at the back the newly trained pwc net recovers the hole between the cart and the front person better than raft abstract how important are training details and datasets to recent optical flow models like raft and do they generalize to explore these questions rather than develop a new model we revisit three prominent models pwc net irr pwc and raft with a common set of modern training techniques and datasets and observe significant performance gains demonstrating the importance and generality of these training details our newly trained pwc net and irr pwc models show surprisingly large improvements up to 30 versus original published results on sintel and kitti 2015 benchmarks they outperform the more recent flow1d on kitti 2015 while being 3 faster during inference our newly trained raft achieves an fl all score of 4 31 on kitti 2015 more accurate than all published optical flow methods at the time of writing our results demonstrate the benefits of separating the contributions of models training techniques and datasets when analyzing performance gains of optical flow methods papers disentangling architecture and training for optical flow deqing sun t charles herrmann fitsum reda michael rubinstein david j fleet and william t freeman eccv 2022 t project lead equal technical contribution arxiv cvf code link to code for disentangling architecture and training for optical flow bibtex inproceedings sun2022disentangling title disentangling architecture and training for optical flow author sun deqing and herrmann charles and reda fitsum and rubinstein michael and fleet david j and freeman william t booktitle eccv year 2022 back to top autoflow learning a better training set for optical flow deqing sun daniel vlasic charles herrmann varun jampani michael krainin huiwen chang ramin zabih william t freeman ce liu google research paper samples code available now dataset available now left pipelines for optical flow a typical pipeline pre trains models on static datasets e g flyingchairs and then evaluates the performance on a target dataset e g sintel autoflow learns pre training data which is optimized ona target dataset right accuracy w r t number of pre training examples on sintel final four autoflow pre training examples with augmentation achieve lower errors than 22 872 flyingchairs pre training examples with augmentation the gap between pwc net and raft becomes small when pre trained on enough autoflow examples abstract synthetic datasets play a critical role in pre training cnn models for optical flow but they are painstaking to generate and hard to adapt to new applications to automate the process we present autoflow a simple and effective method to render training data for optical flow that optimizes the performance of a model on a target dataset autoflow takes a layered approach to render synthetic data where the motion shape and appearance of each layer are controlled by learnable hyperparameters experimental results show that autoflow achieves state of the art accuracy in pre training both pwc net and raft papers autoflow learning a better training set for optical flow deqing sun daniel vlasic charles herrmann varun jampani michael krainin huiwen chang ramin zabih william t freeman and ce liu oral presentation cvpr 2021 arxiv cvf samples code link to code for autoflow learning a better training set for optical flow dataset static dataset with 40 000 training examples part 1 part 2 part 3 part 4 96g in total license cc by bibtex inproceedings sun2021autoflow title autoflow learning a better training set for optical flow author sun deqing and vlasic daniel and herrmann charles and jampani varun and krainin michael and chang huiwen and zabih ramin and freeman william t and liu ce booktitle cvpr year 2021 back to top