Meta tags:

Headings (most frequently used words):

oribench, orientation, grounding, results, additional, origen, zero, shot, 3d, in, text, to, image, generation, method, experimental, reward, single, multi, tl, dr, abstract, qualitative, bibtex, we, introduce, the, first, for, across, multiple, objects, and, diverse, categories, adaptive, time, rescaled, langevin, sde, setup, quantitative, user, study, general, orientations, primitive, views,

Text of the page (most frequently used words):
the (35), and (22), #orientation (20), mathbf (18), image (15), grounding (13), #origen (12), reward (12), mathcal (10), text (9), generation (8), for (8), zero (7), oribench (7), single (7), that (7), object (7), introduce (7), based (7), gamma (7), this (6), images (6), models (6), shot (5), additional (5), multi (5), using (5), big (5), orientations (4), results (4), sqrt (4), hat (4), langevin (4), time (4), model (4), across (4), from (3), arxiv (3), alignment (3), while (3), over (3), which (3), with (3), methods (3), guided (3), metrics (3), diverse (3), prompts (3), gradient (3), ascent (3), approach (3), nabla (3), function (3), adaptive (3), code (3), multiple (3), objects (3), method (3), website (2), yunhong (2), min (2), daehyeon (2), choi (2), kyeongmin (2), yeo (2), jihyun (2), lee (2), minhyuk (2), sung (2), 2025 (2), 2503 (2), 22194 (2), bibtex (2), primitive (2), views (2), general (2), input (2), prompt (2), generated (2), three (2), c3dw (2), were (2), participants (2), both (2), baseline (2), user (2), outperforms (2), vanilla (2), t2i (2), times (2), view (2), setting (2), control (2), unlike (2), trained (2), has (2), realism (2), also (2), requiring (2), may (2), absolute (2), error (2), accuracy (2), evaluation (2), use (2), score (2), quantitative (2), scenario (2), total (2), various (2), azimuths (2), experimental (2), latent (2), representation (2), eta (2), noise (2), epsilon_i (2), convergence (2), step (2), frac (2), rescaling (2), line (2), dynamics (2), problem (2), mathrm (2), crop (2), w_i (2), phi_i (2), estimation (2), first (2), categories (2), sampling (2), guidance (2), adapted, instantdrag, nerfies, licensed, under, creative, commons, attribution, sharealike, international, license, misc, min2025origen, title, author, year, eprint, archiveprefix, primaryclass, url, https, org, abs, qualitative, provided, asked, select, best, reflected, result, was, preferred, outperforming, study, all, flux, schnell, achieves, highest, among, surpasses, highlights, limitations, struggle, precise, due, ambiguity, textual, descriptions, these, consistently, generates, accurately, align, specified, cheng, synthetic, data, learn, thus, limited, generalizability, real, world, output, lack, liu, but, without, backgrounds, background, composition, unnatrual, artifacts, existing, dps, mpgd, freedom, reno, achieve, suboptimal, compared, measure, two, azimuth, angles, between, predicted, angular, within, tolerance, orientanything, predict, case, pickscore, vqa, clip, acc, benchmark, coco, dataset, consist, ground, truth, 1000, 371, more, please, refer, appendix, paper, setup, efficiently, sample, optimal, aligned, distribution, traditional, get, stuck, local, optima, incorporates, stochasticity, leading, following, simple, discretized, update, rule, epsilon_, note, implementation, requires, only, add, gaussian, sim, further, enhance, speed, performance, modifying, process, monitor, dynamically, adjusts, size, log, rescaled, sde, formulate, define, given, target, negative, divergence, follows, sum_, here, orient, anything, extracts, centered, groundingdino, open, set, detection, inherently, supports, averaging, rewards, maximization, previous, work, spatial, primarily, focused, positioning, lacks, address, propose, pretrained, discriminative, one, generative, flow, optimization, natural, choice, struggles, maintain, instead, adopt, extends, simply, injecting, random, just, additionally, accelerate, our, experiments, show, training, test, studies, abstract, neurips, equal, contribution, kaist,

Text of the page (random words):
origen zero shot 3d orientation grounding in text to image generation origen zero shot 3d orientation grounding in text to image generation yunhong min daehyeon choi kyeongmin yeo jihyun lee minhyuk sung kaist equal contribution neurips 2025 arxiv code bibtex tl dr we introduce origen the first zero shot method for 3d orientation grounding in text to image generation across multiple objects and diverse categories abstract we introduce origen the first zero shot method for 3d orientation grounding in text to image generation across multiple objects and diverse categories while previous work on spatial grounding in image generation has primarily focused on 2d positioning it lacks control over 3d orientation to address this we propose a reward guided sampling approach using a pretrained discriminative model for 3d orientation estimation and a one step text to image generative flow model while gradient ascent based optimization is a natural choice for reward based guidance it struggles to maintain image realism instead we adopt a sampling based approach using langevin dynamics which extends gradient ascent by simply injecting random noise requiring just a single additional line of code additionally we introduce adaptive time rescaling based on the reward function to accelerate convergence our experiments show that origen outperforms both training based and test time guidance methods across quantitative metrics and user studies method orientation grounding reward to formulate the orientation grounding problem as a reward maximization problem we define the orientation grounding reward for a given target 3d orientation pi phi_i and image mathbf i using negative kl divergence as follows mathcal r mathbf i frac 1 n sum_ i 1 n d_ text kl big mathcal d big mathrm crop mathbf i w_i big big pi phi_i big here mathcal d is the orientation estimation model orient anything and mathrm crop mathbf i w_i extracts a centered object image using groundingdino an open set object detection model this reward function inherently supports multi object orientation grounding by averaging rewards across multiple objects n reward adaptive time rescaled langevin sde we introduce reward guided langevin dynamics to efficiently sample a latent representation mathbf x from the optimal reward aligned distribution unlike traditional gradient ascent which may get stuck in local optima this approach incorporates stochasticity leading to the following simple discretized update rule mathbf x _ i 1 sqrt 1 gamma mathbf x _i gamma eta nabla hat mathcal r mathbf x _i sqrt gamma epsilon_ i note that for implementation this requires only a single line of code to add gaussian noise epsilon_i sim mathcal n 0 mathbf i to the latent representation to further enhance convergence speed and performance we introduce reward adaptive time rescaling modifying the langevin process with a monitor function mathcal g hat mathcal r mathbf x that dynamically adjusts the step size based on the reward mathbf x _ i 1 sqrt 1 gamma mathbf x _i mathbf x _i gamma mathbf x _i eta nabla hat mathcal r mathbf x _i frac 1 2 gamma mathbf x _i nabla log mathcal g hat mathcal r mathbf x _i sqrt gamma mathbf x _i epsilon_i experimental results experimental setup we introduce oribench benchmark based on the ms coco dataset that consist of diverse text prompts image and ground truth orientations 1 oribench single single object scenario total 1000 prompts various azimuths 2 oribench multi multi object scenario total 371 prompts various azimuths for more evaluation general orientation primitive views please refer to the appendix of the paper quantitative results we measure orientation grounding accuracy using two metrics 1 absolute error the absolute error on azimuth angles between the predicted and grounding object orientations 2 acc 22 5 the angular accuracy within a tolerance of 22 5 for evaluation we use orientanything to predict the 3d orientation from the generated images in the case of text image alignment we use three metrics clip score vqa score and pickscore oribench single c3dw cheng et al is trained on synthetic data to learn orientation to image generation thus it has limited generalizability to real world images and the output images lack realism zero 1 to 3 liu et al is also trained on single object images but without backgrounds requiring additional background image composition that may introduce unnatrual artifacts the existing methods dps mpgd freedom reno on guided generation methods also achieve suboptimal results compared to origen oribench multi origen outperforms all baseline models in orientation alignment while flux schnell achieves the highest alignment among vanilla t2i models origen surpasses it by over 2 5 times in the 3 view setting 82 4 vs 31 2 and over 2 times in the 4 view setting 86 6 vs 42 4 this highlights the limitations of vanilla t2i models which struggle with precise orientation control due to the ambiguity in textual descriptions unlike these models origen consistently generates images that accurately align with the specified orientations user study the input prompt and images generated by three models zero 1 to 3 c3dw and origen were provided and participants were asked to select the image that best reflected both the input prompt and the grounding orientation as a result origen was preferred by 58 18 of the participants outperforming the baseline models additional qualitative results oribench single oribench multi additional general orientations additional primitive views bibtex misc min2025origen title origen zero shot 3d orientation grounding in text to image generation author yunhong min and daehyeon choi and kyeongmin yeo and jihyun lee and minhyuk sung year 2025 eprint 2503 22194 archiveprefix arxiv primaryclass cs cv url https arxiv org abs 2503 22194 this website is licensed under a creative commons attribution sharealike 4 0 international license website adapted from nerfies and instantdrag