Karim Bouyarmane

Wenyi Wu, Karim Bouyarmane, Ismail Tutar

We present Catalog Phrase Grounding (CPG), a model that can associate product textual data (title, brands) into corresponding regions of product images (isolated product region, brand logo region) for e-commerce vision-language applications. We use a state-of-the-art modulated multimodal transformer encoder-decoder architecture unifying object detection and phrase-grounding. We train the model in self-supervised fashion with 2.3 million image-text pairs synthesized from an e-commerce site. The self-supervision data is annotated with high-confidence pseudo-labels generated with a combination of teacher models: a pre-trained general domain phrase grounding model (e.g. MDETR) and a specialized logo detection model. This allows CPG, as a student model, to benefit from transfer knowledge from these base models combining general-domain knowledge and specialized knowledge. Beyond immediate catalog phrase grounding tasks, we can benefit from CPG representations by incorporating them as ML features into downstream catalog applications that require deep semantic understanding of products. Our experiments on product-brand matching, a challenging e-commerce application, show that incorporating CPG representations into the existing production ensemble system leads to on average 5% recall improvement across all countries globally (with the largest lift of 11% in a single country) at fixed 95% precision, outperforming other alternatives including a logo detection teacher model and ResNet50.

Wisdom Ikezogwo, Mehmet Saygin Seyfioglu, Ranjay Krishna et al.

Reinforcement learning with verifiable rewards (RLVR) and Rubrics as Rewards (RaR) have driven strong gains in domains with clear correctness signals and even in subjective domains by synthesizing evaluation criteria from ideal reference answers. But many real-world tasks admit multiple valid outputs and lack the single ideal answer that rubric generation depends on. We identify this reference-free setting as a gap in current post-training methods and propose Implicit Error Counting (IEC) to fill it. Instead of checking what a response gets right against a rubric, IEC enumerates what it gets wrong, applying severity-weighted scores across task-relevant axes and converting them into calibrated per-aspect rewards. We show that naïve explicit enumeration is too noisy for stable optimization, and that two design choices: implicit score emission and group calibration are necessary to make error counting a reliable reward. As a case study, we validate IEC on virtual try-on (VTO), a domain that is simultaneously too constrained for holistic scoring and too permissive for rubric-based evaluation: subtle garment errors are unacceptable, yet many output variations are correct. We introduce Cascaded Error Counting (CEC) as an evaluation metric, which tracks human preferences well (60% top-1 vs. 30% others), and curate Mismatch-DressCode (MDressBench), a benchmark with maximal attribute mismatch to stress-test reward designs. On MDressBench, IEC outperforms RaR across all metrics (CEC: 5.31 vs. 5.60 on flat references; 5.20 vs. 5.53 on non-flat). On VITON-HD and DressCode, IEC matches or surpasses six baselines on 6 of 8 perceptual metrics. These results suggest that when ideal answers are unavailable, counting errors provide a stronger signal than constructing rubrics.

Qi Li, Shuwen Qiu, Julien Han et al.

The rapid growth of e-commerce has intensified the demand for Virtual Try-On (VTO) technologies, enabling customers to realistically visualize products overlaid on their own images. Despite recent advances, existing VTO models face challenges with fine-grained detail preservation, robustness to real-world imagery, efficient sampling, image editing capabilities, and generalization across diverse product categories. In this paper, we present DiT-VTON, a novel VTO framework that leverages a Diffusion Transformer (DiT), renowned for its performance on text-conditioned image generation, adapted here for the image-conditioned VTO task. We systematically explore multiple DiT configurations, including in-context token concatenation, channel concatenation, and ControlNet integration, to determine the best setup for VTO image conditioning. To enhance robustness, we train the model on an expanded dataset encompassing varied backgrounds, unstructured references, and non-garment categories, demonstrating the benefits of data scaling for VTO adaptability. DiT-VTON also redefines the VTO task beyond garment try-on, offering a versatile Virtual Try-All (VTA) solution capable of handling a wide range of product categories and supporting advanced image editing functionalities such as pose preservation, localized editing, texture transfer, and object-level customization. Experimental results show that our model surpasses state-of-the-art methods on VITON-HD, achieving superior detail preservation and robustness without reliance on additional condition encoders. It also outperforms models with VTA and image editing capabilities on a diverse dataset spanning thousands of product categories.

Julien Han, Shuwen Qiu, Qi Li et al.

We present InstructVTON, an instruction-following interactive virtual try-on system that allows fine-grained and complex styling control of the resulting generation, guided by natural language, on single or multiple garments. A computationally efficient and scalable formulation of virtual try-on formulates the problem as an image-guided or image-conditioned inpainting task. These inpainting-based virtual try-on models commonly use a binary mask to control the generation layout. Producing a mask that yields desirable result is difficult, requires background knowledge, might be model dependent, and in some cases impossible with the masking-based approach (e.g. trying on a long-sleeve shirt with "sleeves rolled up" styling on a person wearing long-sleeve shirt with sleeves down, where the mask will necessarily cover the entire sleeve). InstructVTON leverages Vision Language Models (VLMs) and image segmentation models for automated binary mask generation. These masks are generated based on user-provided images and free-text style instructions. InstructVTON simplifies the end-user experience by removing the necessity of a precisely drawn mask, and by automating execution of multiple rounds of image generation for try-on scenarios that cannot be achieved with masking-based virtual try-on models alone. We show that InstructVTON is interoperable with existing virtual try-on models to achieve state-of-the-art results with styling control.

Xingzi Xu, Qi Li, Shuwen Qiu et al.

Diffusion models enable high-quality virtual try-on (VTO) with their established image synthesis abilities. Despite the extensive end-to-end training of large pre-trained models involved in current VTO methods, real-world applications often prioritize limited training and inference, serving, and deployment budgets for VTO. To solve this obstacle, we apply Doob's h-transform efficient fine-tuning (DEFT) for adapting large pre-trained unconditional models for downstream image-conditioned VTO abilities. DEFT freezes the pre-trained model's parameters and trains a small h-transform network to learn a conditional h-transform. The h-transform network allows training only 1.42 percent of the frozen parameters, compared to a baseline of 5.52 percent in traditional parameter-efficient fine-tuning (PEFT). To further improve DEFT's performance and decrease existing models' inference time, we additionally propose an adaptive consistency loss. Consistency training distills slow but high-performing diffusion models into a fast one while retaining performance by enforcing consistencies along the inference path. Inspired by constrained optimization, instead of distillation, we combine the consistency loss and the denoising score matching loss in a data-adaptive manner for fine-tuning existing VTO models at a low cost. Empirical results show the proposed DEFT-VTON method achieves state-of-the-art performance on VTO tasks, with as few as 15 denoising steps, while maintaining competitive results.

Amir Tavanaei, Kee Kiat Koo, Hayreddin Ceker et al.

In this paper, we study the problem of generating structured objects that conform to a complex schema, with intricate dependencies between the different components (facets) of the object. The facets of the object (attributes, fields, columns, properties) can be a mix of short, structured, type-constrained facts, or long natural-language descriptions. The object has to be self-consistent between the different facets in the redundant information it carries (relative consistency), while being grounded with respect to world knowledge (absolute consistency). We frame the problem as a Language Modeling problem (Structured Object Language Modeling) and train an LLM to perform the task natively, without requiring instructions or prompt-engineering. We propose a self-supervised denoising method to train the model from an existing dataset of such objects. The input query can be the existing object itself, in which case the model acts as a regenerator, completing, correcting, normalizing the input, or any unstructured blurb to be structured. We show that the self-supervised denoising training provides a strong baseline, and that additional supervised fine-tuning with small amount of human demonstrations leads to further improvement. Experimental results show that the proposed method matches or outperforms prompt-engineered general-purpose state-of-the-art LLMs (Claude 3, Mixtral-8x7B), while being order-of-magnitude more cost-efficient.

Qi Li, Shuwen Qiu, Julien Han et al.

As online shopping continues to grow, the demand for Virtual Try-On (VTON) technology has surged, allowing customers to visualize products on themselves by overlaying product images onto their own photos. An essential yet challenging condition for effective VTON is pose control, which ensures accurate alignment of products with the user's body while supporting diverse orientations for a more immersive experience. However, incorporating pose conditions into VTON models presents several challenges, including selecting the optimal pose representation, integrating poses without additional parameters, and balancing pose preservation with flexible pose control. In this work, we build upon a baseline VTON model that concatenates the reference image condition without external encoder, control network, or complex attention layers. We investigate methods to incorporate pose control into this pure concatenation paradigm by spatially concatenating pose data, comparing performance using pose maps and skeletons, without adding any additional parameters or module to the baseline model. Our experiments reveal that pose stitching with pose maps yields the best results, enhancing both pose preservation and output realism. Additionally, we introduce a mixed-mask training strategy using fine-grained and bounding box masks, allowing the model to support flexible product integration across varied poses and conditions.

Mengqi Zhang, Qi Li, Mehmet Saygin Seyfioglu et al.

Virtual Try-On is a promising research area with broad applications in e-commerce and everyday life, enabling users to visualize garments on themselves or others before purchase. Most existing methods depend on predefined or user-specified masks to guide garment placement, but their performance is highly sensitive to mask quality, often causing misalignment or artifacts, and introduces redundant steps for users. To overcome these limitations, we propose a mask-free virtual try-on framework that requires only minimal modifications to the underlying architecture while remaining compatible with common diffusion-based pipelines. To address the increased ambiguity in the absence of masks, we integrate an attention-based guidance mechanism that explicitly directs the model to focus on the target garment region and improves correspondence between the garment and the person. Additionally, we incorporate instruction prompts, allowing users to flexibly control garment categories and wearing styles, addressing the underutilization of prompts in prior work and improving interaction flexibility. Both qualitative and quantitative evaluations across multiple datasets demonstrate that our approach consistently outperforms existing methods, producing more accurate, robust, and user-friendly try-on results.

Alex Ayoub, Kavosh Asadi, Dale Schuurmans et al.

Large reasoning models (LRMs) often consume excessive tokens, inflating computational cost and latency. We challenge the assumption that longer responses improve accuracy. By penalizing reasoning tokens using a discounted reinforcement learning setup (interpretable as a small token cost) and analyzing Blackwell optimality in restricted policy classes, we encourage concise yet accurate reasoning. Experiments confirm our theoretical results that this approach shortens chains of thought while preserving accuracy.

Ege Beyazit, KL Navaneet, Prashant Mathur et al.

Black-box Large Language Models (LLMs) provide practical and accessible alternatives to other machine learning methods, as they require minimal labeled data and machine learning expertise to develop solutions for various decision making problems. However, for applications that need operating with constraints on specific metrics (e.g., precision $\geq$ 95%), decision making with black-box LLMs remains unfavorable, due to their low numerical output cardinalities. This results in limited control over their operating points, preventing fine-grained adjustment of their decision making behavior. In this paper, we study using black-box LLMs as classifiers, focusing on efficiently improving their operational granularity without performance loss. Specifically, we first investigate the reasons behind their low-cardinality numerical outputs and show that they are biased towards generating rounded but informative verbalized probabilities. Then, we experiment with standard prompt engineering, uncertainty estimation and confidence elicitation techniques, and observe that they do not effectively improve operational granularity without sacrificing performance or increasing inference cost. Finally, we propose efficient approaches to significantly increase the number and diversity of available operating points. Our proposed approaches provide finer-grained operating points and achieve comparable to or better performance than the benchmark methods across 11 datasets and 3 LLMs.

Kavosh Asadi, Julien Han, Idan Pipano et al.

Direct preference optimization (\texttt{DPO}) has emerged as a promising approach for solving the alignment problem in AI. In this paper, we make two counter-intuitive observations about \texttt{DPO}. First, we show that \texttt{DPO} loss could be derived by starting from an alternative optimization problem that only defines the KL guardrail on in-sample responses, unlike the original RLHF problem where guardrails are defined on the entire distribution. Second, we prove a surprising property of this alternative optimization problem, namely that under its optimal policy, both preferred and rejected responses tend to decrease in probability, a phenomenon typically displayed by DPO in practice. To control this behavior, we propose a set of constraints designed to limit the displacement of probability mass between the preferred and rejected responses in the reference and target policies. The resulting algorithm, which we call Constrained Controlled DPO (\texttt{C2-DPO}), has a meaningful RLHF interpretation. By hedging against the displacement, \texttt{C2-DPO} provides practical improvements over vanilla \texttt{DPO} when aligning several language models using standard preference datasets.

Xingzi Xu, Amir Tavanaei, Kavosh Asadi et al.

Despite very fast progress, efficiently training large language models (LLMs) in very long contexts remains challenging. Existing methods fall back to training LLMs with short contexts (a maximum of a few thousands tokens in training) and use inference time techniques when evaluating on long contexts (above 1M tokens context window at inference). As opposed to long-context-inference, training on very long context input prompts is quickly limited by GPU memory availability and by the prohibitively long training times it requires on state-of-the-art hardware. Meanwhile, many real-life applications require not only inference but also training/fine-tuning with long context on specific tasks. Such applications include, for example, augmenting the context with various sources of raw reference information for fact extraction, fact summarization, or fact reconciliation tasks. We propose adjoint sharding, a novel technique that comprises sharding gradient calculation during training to reduce memory requirements by orders of magnitude, making training on very long context computationally tractable. Adjoint sharding is based on the adjoint method and computes equivalent gradients to backpropagation. We also propose truncated adjoint sharding to speed up the algorithm while maintaining performance. We provide a distributed version, and a paralleled version of adjoint sharding to further speed up training. Empirical results show the proposed adjoint sharding algorithm reduces memory usage by up to 3X with a 1.27B parameter large language model on 1M context length training. This allows to increase the maximum context length during training or fine-tuning of a 1.27B parameter model from 35K tokens to above 100K tokens on a training infrastructure composed of five AWS P4 instances.

Wenliang Zhong, Wenyi Wu, Qi Li et al.

Multimodal Large Language Models (MLLMs) have achieved SOTA performance in various visual language tasks by fusing the visual representations with LLMs leveraging some visual adapters. In this paper, we first establish that adapters using query-based Transformers such as Q-former is a simplified Multi-instance Learning method without considering instance heterogeneity/correlation. We then propose a general component termed Multi-instance Visual Prompt Generator (MIVPG) to incorporate enriched visual representations into LLMs by taking advantage of instance correlation between images or patches for the same sample. Quantatitive evaluation on three public vision-language (VL) datasets from different scenarios shows that the proposed MIVPG improves Q-former in main VL tasks.

Mehmet Saygin Seyfioglu, Karim Bouyarmane, Suren Kumar et al.

As online shopping is growing, the ability for buyers to virtually visualize products in their settings-a phenomenon we define as "Virtual Try-All"-has become crucial. Recent diffusion models inherently contain a world model, rendering them suitable for this task within an inpainting context. However, traditional image-conditioned diffusion models often fail to capture the fine-grained details of products. In contrast, personalization-driven models such as DreamPaint are good at preserving the item's details but they are not optimized for real-time applications. We present "Diffuse to Choose," a novel diffusion-based image-conditioned inpainting model that efficiently balances fast inference with the retention of high-fidelity details in a given reference item while ensuring accurate semantic manipulations in the given scene content. Our approach is based on incorporating fine-grained features from the reference image directly into the latent feature maps of the main diffusion model, alongside with a perceptual loss to further preserve the reference item's details. We conduct extensive testing on both in-house and publicly available datasets, and show that Diffuse to Choose is superior to existing zero-shot diffusion inpainting methods as well as few-shot diffusion personalization algorithms like DreamPaint.

Mehmet Saygin Seyfioglu, Karim Bouyarmane, Suren Kumar et al.

We introduce DreamPaint, a framework to intelligently inpaint any e-commerce product on any user-provided context image. The context image can be, for example, the user's own image for virtual try-on of clothes from the e-commerce catalog on themselves, the user's room image for virtual try-on of a piece of furniture from the e-commerce catalog in their room, etc. As opposed to previous augmented-reality (AR)-based virtual try-on methods, DreamPaint does not use, nor does it require, 3D modeling of neither the e-commerce product nor the user context. Instead, it directly uses 2D images of the product as available in product catalog database, and a 2D picture of the context, for example taken from the user's phone camera. The method relies on few-shot fine tuning a pre-trained diffusion model with the masked latents (e.g., Masked DreamBooth) of the catalog images per item, whose weights are then loaded on a pre-trained inpainting module that is capable of preserving the characteristics of the context image. DreamPaint allows to preserve both the product image and the context (environment/user) image without requiring text guidance to describe the missing part (product/context). DreamPaint also allows to intelligently infer the best 3D angle of the product to place at the desired location on the user context, even if that angle was previously unseen in the product's reference 2D images. We compare our results against both text-guided and image-guided inpainting modules and show that DreamPaint yields superior performance in both subjective human study and quantitative metrics.