arXiv subjects

Gerard Medioni

Publications and source records attributed to Gerard Medioni.

At least 19 recordsLinked to original sources

arXiv2026

CtrlSpeech: Coarse-to-Fine Control for Expressive Speech Synthesis

Recent Text-To-Speech (TTS) systems have achieved strong naturalness and zero-shot voice cloning performance, but fine-grained control of expressive speech at the word or phoneme level remains challenging. We propose CtrlSpeech, a controllable, expressive TTS framework with coarse-to-fine control. Built on the DiTAR architecture, CtrlSpeech combines global speaker conditioning with phone-aligned pitch, loudness, and duration signals, enabling localized prosodic control while preserving the target speaker's timbre. This design allows users to adjust expressive attributes at a fine temporal granularity, making speech refinement more flexible and controllable. Experimental results show that CtrlSpeech achieves competitive zero-shot TTS performance and improves controllability over expressive attributes, demonstrating its effectiveness for flexible and practical expressive speech synthesis.

eess.AS↗

arXiv2025

RosettaSpeech: Zero-Shot Speech-to-Speech Translation without Parallel Speech

End-to-end speech-to-speech translation (S2ST) systems typically struggle with a critical data bottleneck: the scarcity of parallel speech-to-speech corpora. To overcome this, we introduce RosettaSpeech, a novel zero-shot framework trained exclusively on monolingual speech-text data augmented by machine translation supervision. Unlike prior works that rely on complex cascaded pseudo-labeling, our approach strategically utilizes text as a semantic bridge during training to synthesize translation targets, thereby eliminating the need for parallel speech pairs while maintaining a direct, end-to-end inference pipeline. Empirical evaluations on the CVSS-C benchmark demonstrate that RosettaSpeech achieves state-of-the-art zero-shot performance, surpassing leading baselines by significant margins - achieving ASR-BLEU scores of 25.17 for German-to-English (+27% relative gain) and 29.86 for Spanish-to-English (+14%). Crucially, our model effectively preserves the source speaker's voice without ever seeing paired speech data. We further analyze the impact of data scaling and demonstrate the model's capability in many-to-one translation, offering a scalable solution for extending high-quality S2ST to "text-rich, speech-poor" languages.

eess.AS↗

arXiv2025

Group-Aware Reinforcement Learning for Output Diversity in Large Language Models

Large Language Models (LLMs) often suffer from mode collapse, repeatedly generating the same few completions even when many valid answers exist, limiting their diversity across a wide range of tasks. We introduce Group-Aware Policy Optimization (GAPO), a simple extension of the recent and popular Group Relative Policy Optimization (GRPO) that computes rewards over the group as a whole. GAPO enables learning from the group-level properties such as diversity and coverage. We demonstrate GAPO using a frequency-aware reward function that encourages uniform sampling over valid LLM completions, and show that GAPO-trained models produce valid and more diverse model responses. Beyond this setup, GAPO generalizes to open-ended prompts and improves response diversity without compromising accuracy on standard LLM benchmarks (GSM8K, MATH, HumanEval, MMLU-Pro). Our code will be made publicly available.

cs.CL↗

arXiv2025

LV-MAE: Learning Long Video Representations through Masked-Embedding Autoencoders

In this work, we introduce long-video masked-embedding autoencoders (LV-MAE), a self-supervised learning framework for long video representation. Our approach treats short- and long-span dependencies as two separate tasks. Such decoupling allows for a more intuitive video processing where short-span spatiotemporal primitives are first encoded and are then used to capture long-range dependencies across consecutive video segments. To achieve this, we leverage advanced off-the-shelf multimodal encoders to extract representations from short segments within the long video, followed by pre-training a masked-embedding autoencoder capturing high-level interactions across segments. LV-MAE is highly efficient to train and enables the processing of much longer videos by alleviating the constraint on the number of input frames. Furthermore, unlike existing methods that typically pre-train on short-video datasets, our approach offers self-supervised pre-training using long video samples (e.g., 20+ minutes video clips) at scale. Using LV-MAE representations, we achieve state-of-the-art results on three long-video benchmarks -- LVU, COIN, and Breakfast -- employing only a simple classification head for either attentive or linear probing. Finally, to assess LV-MAE pre-training and visualize its reconstruction quality, we leverage the video-language aligned space of short video representations to monitor LV-MAE through video-text retrieval. Code is available at https://github.com/amazon-science/lv-mae.

cs.CV↗

arXiv2024

Data Pruning via Separability, Integrity, and Model Uncertainty-Aware Importance Sampling

This paper improves upon existing data pruning methods for image classification by introducing a novel pruning metric and pruning procedure based on importance sampling. The proposed pruning metric explicitly accounts for data separability, data integrity, and model uncertainty, while the sampling procedure is adaptive to the pruning ratio and considers both intra-class and inter-class separation to further enhance the effectiveness of pruning. Furthermore, the sampling method can readily be applied to other pruning metrics to improve their performance. Overall, the proposed approach scales well to high pruning ratio and generalizes better across different classification models, as demonstrated by experiments on four benchmark datasets, including the fine-grained classification scenario.

cs.CV↗

arXiv2023

Asymmetric Image Retrieval with Cross Model Compatible Ensembles

The asymmetrical retrieval setting is a well suited solution for resource constrained applications such as face recognition and image retrieval. In this setting, a large model is used for indexing the gallery while a lightweight model is used for querying. The key principle in such systems is ensuring that both models share the same embedding space. Most methods in this domain are based on knowledge distillation. While useful, they suffer from several drawbacks: they are upper-bounded by the performance of the single best model found and cannot be extended to use an ensemble of models in a straightforward manner. In this paper we present an approach that does not rely on knowledge distillation, rather it utilizes embedding transformation models. This allows the use of N independently trained and diverse gallery models (e.g., trained on different datasets or having a different architecture) and a single query model. As a result, we improve the overall accuracy beyond that of any single model while maintaining a low computational budget for querying. Additionally, we propose a gallery image rejection method that utilizes the diversity between multiple transformed embeddings to estimate the uncertainty of gallery images.

cs.CV↗

arXiv2023

FPGAN-Control: A Controllable Fingerprint Generator for Training with Synthetic Data

Training fingerprint recognition models using synthetic data has recently gained increased attention in the biometric community as it alleviates the dependency on sensitive personal data. Existing approaches for fingerprint generation are limited in their ability to generate diverse impressions of the same finger, a key property for providing effective data for training recognition models. To address this gap, we present FPGAN-Control, an identity preserving image generation framework which enables control over the fingerprint's image appearance (e.g., fingerprint type, acquisition device, pressure level) of generated fingerprints. We introduce a novel appearance loss that encourages disentanglement between the fingerprint's identity and appearance properties. In our experiments, we used the publicly available NIST SD302 (N2N) dataset for training the FPGAN-Control model. We demonstrate the merits of FPGAN-Control, both quantitatively and qualitatively, in terms of identity preservation level, degree of appearance control, and low synthetic-to-real domain gap. Finally, training recognition models using only synthetic datasets generated by FPGAN-Control lead to recognition accuracies that are on par or even surpass models trained using real data. To the best of our knowledge, this is the first work to demonstrate this.

cs.CV↗

arXiv2023

Synthetic Data for Model Selection

Recent breakthroughs in synthetic data generation approaches made it possible to produce highly photorealistic images which are hardly distinguishable from real ones. Furthermore, synthetic generation pipelines have the potential to generate an unlimited number of images. The combination of high photorealism and scale turn synthetic data into a promising candidate for improving various machine learning (ML) pipelines. Thus far, a large body of research in this field has focused on using synthetic images for training, by augmenting and enlarging training data. In contrast to using synthetic data for training, in this work we explore whether synthetic data can be beneficial for model selection. Considering the task of image classification, we demonstrate that when data is scarce, synthetic data can be used to replace the held out validation set, thus allowing to train on a larger dataset. We also introduce a novel method to calibrate the synthetic error estimation to fit that of the real domain. We show that such calibration significantly improves the usefulness of synthetic data for model selection.

cs.CV↗

arXiv2023

Shape of You: Precise 3D shape estimations for diverse body types

This paper presents Shape of You (SoY), an approach to improve the accuracy of 3D body shape estimation for vision-based clothing recommendation systems. While existing methods have successfully estimated 3D poses, there remains a lack of work in precise shape estimation, particularly for diverse human bodies. To address this gap, we propose two loss functions that can be readily integrated into parametric 3D human reconstruction pipelines. Additionally, we propose a test-time optimization routine that further improves quality. Our method improves over the recent SHAPY method by 17.7% on the challenging SSP-3D dataset. We consider our work to be a step towards a more accurate 3D shape estimation system that works reliably on diverse body types and holds promise for practical applications in the fashion industry.

cs.CV↗

arXiv2023

GPT4Rec: A Generative Framework for Personalized Recommendation and User Interests Interpretation

Recent advancements in Natural Language Processing (NLP) have led to the development of NLP-based recommender systems that have shown superior performance. However, current models commonly treat items as mere IDs and adopt discriminative modeling, resulting in limitations of (1) fully leveraging the content information of items and the language modeling capabilities of NLP models; (2) interpreting user interests to improve relevance and diversity; and (3) adapting practical circumstances such as growing item inventories. To address these limitations, we present GPT4Rec, a novel and flexible generative framework inspired by search engines. It first generates hypothetical "search queries" given item titles in a user's history, and then retrieves items for recommendation by searching these queries. The framework overcomes previous limitations by learning both user and item embeddings in the language space. To well-capture user interests with different aspects and granularity for improving relevance and diversity, we propose a multi-query generation technique with beam search. The generated queries naturally serve as interpretable representations of user interests and can be searched to recommend cold-start items. With GPT-2 language model and BM25 search engine, our framework outperforms state-of-the-art methods by $75.7\%$ and $22.2\%$ in Recall@K on two public datasets. Experiments further revealed that multi-query generation with beam search improves both the diversity of retrieved items and the coverage of a user's multi-interests. The adaptiveness and interpretability of generated queries are discussed with qualitative case studies.

cs.IR↗

arXiv2022

Contrastive Learning for Interactive Recommendation in Fashion

Recommender systems and search are both indispensable in facilitating personalization and ease of browsing in online fashion platforms. However, the two tools often operate independently, failing to combine the strengths of recommender systems to accurately capture user tastes with search systems' ability to process user queries. We propose a novel remedy to this problem by automatically recommending personalized fashion items based on a user-provided text request. Our proposed model, WhisperLite, uses contrastive learning to capture user intent from natural language text and improves the recommendation quality of fashion products. WhisperLite combines the strength of CLIP embeddings with additional neural network layers for personalization, and is trained using a composite loss function based on binary cross entropy and contrastive loss. The model demonstrates a significant improvement in offline recommendation retrieval metrics when tested on a real-world dataset collected from an online retail fashion store, as well as widely used open-source datasets in different e-commerce domains, such as restaurants, movies and TV shows, clothing and shoe reviews. We additionally conduct a user study that captures user judgements on the relevance of the model's recommended items, confirming the relevancy of WhisperLite's recommendations in an online setting.

cs.IR↗

arXiv2022

Identity Preserving Loss for Learned Image Compression

Deep learning model inference on embedded devices is challenging due to the limited availability of computation resources. A popular alternative is to perform model inference on the cloud, which requires transmitting images from the embedded device to the cloud. Image compression techniques are commonly employed in such cloud-based architectures to reduce transmission latency over low bandwidth networks. This work proposes an end-to-end image compression framework that learns domain-specific features to achieve higher compression ratios than standard HEVC/JPEG compression techniques while maintaining accuracy on downstream tasks (e.g., recognition). Our framework does not require fine-tuning of the downstream task, which allows us to drop-in any off-the-shelf downstream task model without retraining. We choose faces as an application domain due to the ready availability of datasets and off-the-shelf recognition models as representative downstream tasks. We present a novel Identity Preserving Reconstruction (IPR) loss function which achieves Bits-Per-Pixel (BPP) values that are ~38% and ~42% of CRF-23 HEVC compression for LFW (low-resolution) and CelebA-HQ (high-resolution) datasets, respectively, while maintaining parity in recognition accuracy. The superior compression ratio is achieved as the model learns to retain the domain-specific features (e.g., facial features) while sacrificing details in the background. Furthermore, images reconstructed by our proposed compression model are robust to changes in downstream model architectures. We show at-par recognition performance on the LFW dataset with an unseen recognition model while retaining a lower BPP value of ~38% of CRF-23 HEVC compression.

cs.CV↗

arXiv2022

OutfitTransformer: Learning Outfit Representations for Fashion Recommendation

Learning an effective outfit-level representation is critical for predicting the compatibility of items in an outfit, and retrieving complementary items for a partial outfit. We present a framework, OutfitTransformer, that uses the proposed task-specific tokens and leverages the self-attention mechanism to learn effective outfit-level representations encoding the compatibility relationships between all items in the entire outfit for addressing both compatibility prediction and complementary item retrieval tasks. For compatibility prediction, we design an outfit token to capture a global outfit representation and train the framework using a classification loss. For complementary item retrieval, we design a target item token that additionally takes the target item specification (in the form of a category or text description) into consideration. We train our framework using a proposed set-wise outfit ranking loss to generate a target item embedding given an outfit, and a target item specification as inputs. The generated target item embedding is then used to retrieve compatible items that match the rest of the outfit. Additionally, we adopt a pre-training approach and a curriculum learning strategy to improve retrieval performance. Since our framework learns at an outfit-level, it allows us to learn a single embedding capturing higher-order relations among multiple items in the outfit more effectively than pairwise methods. Experiments demonstrate that our approach outperforms state-of-the-art methods on compatibility prediction, fill-in-the-blank, and complementary item retrieval tasks. We further validate the quality of our retrieval results with a user study.

cs.CV↗

arXiv2022

Efficient Video Instance Segmentation via Tracklet Query and Proposal

Video Instance Segmentation (VIS) aims to simultaneously classify, segment, and track multiple object instances in videos. Recent clip-level VIS takes a short video clip as input each time showing stronger performance than frame-level VIS (tracking-by-segmentation), as more temporal context from multiple frames is utilized. Yet, most clip-level methods are neither end-to-end learnable nor real-time. These limitations are addressed by the recent VIS transformer (VisTR) which performs VIS end-to-end within a clip. However, VisTR suffers from long training time due to its frame-wise dense attention. In addition, VisTR is not fully end-to-end learnable in multiple video clips as it requires a hand-crafted data association to link instance tracklets between successive clips. This paper proposes EfficientVIS, a fully end-to-end framework with efficient training and inference. At the core are tracklet query and tracklet proposal that associate and segment regions-of-interest (RoIs) across space and time by an iterative query-video interaction. We further propose a correspondence learning that makes tracklets linking between clips end-to-end learnable. Compared to VisTR, EfficientVIS requires 15x fewer training epochs while achieving state-of-the-art accuracy on the YouTube-VIS benchmark. Meanwhile, our method enables whole video instance segmentation in a single end-to-end pass without data association at all.

cs.CV↗

arXiv2021

MILA: Multi-Task Learning from Videos via Efficient Inter-Frame Attention

Prior work in multi-task learning has mainly focused on predictions on a single image. In this work, we present a new approach for multi-task learning from videos via efficient inter-frame local attention (MILA). Our approach contains a novel inter-frame attention module which allows learning of task-specific attention across frames. We embed the attention module in a ``slow-fast'' architecture, where the slower network runs on sparsely sampled keyframes and the light-weight shallow network runs on non-keyframes at a high frame rate. We also propose an effective adversarial learning strategy to encourage the slow and fast network to learn similar features. Our approach ensures low-latency multi-task learning while maintaining high quality predictions. Experiments show competitive accuracy compared to state-of-the-art on two multi-task learning benchmarks while reducing the number of floating point operations (FLOPs) by up to 70\%. In addition, our attention based feature propagation method (ILA) outperforms prior work in terms of task accuracy while also reducing up to 90\% of FLOPs.

cs.CV↗

arXiv2021

GAN-Control: Explicitly Controllable GANs

We present a framework for training GANs with explicit control over generated images. We are able to control the generated image by settings exact attributes such as age, pose, expression, etc. Most approaches for editing GAN-generated images achieve partial control by leveraging the latent space disentanglement properties, obtained implicitly after standard GAN training. Such methods are able to change the relative intensity of certain attributes, but not explicitly set their values. Recently proposed methods, designed for explicit control over human faces, harness morphable 3D face models to allow fine-grained control capabilities in GANs. Unlike these methods, our control is not constrained to morphable 3D face model parameters and is extendable beyond the domain of human faces. Using contrastive learning, we obtain GANs with an explicitly disentangled latent space. This disentanglement is utilized to train control-encoders mapping human-interpretable inputs to suitable latent vectors, thus allowing explicit control. In the domain of human faces we demonstrate control over identity, age, pose, expression, hair color and illumination. We also demonstrate control capabilities of our framework in the domains of painted portraits and dog image generation. We demonstrate that our approach achieves state-of-the-art performance both qualitatively and quantitatively.

cs.CV↗

arXiv2021

Energy-Based Learning for Scene Graph Generation

Traditional scene graph generation methods are trained using cross-entropy losses that treat objects and relationships as independent entities. Such a formulation, however, ignores the structure in the output space, in an inherently structured prediction problem. In this work, we introduce a novel energy-based learning framework for generating scene graphs. The proposed formulation allows for efficiently incorporating the structure of scene graphs in the output space. This additional constraint in the learning framework acts as an inductive bias and allows models to learn efficiently from a small number of labels. We use the proposed energy-based framework to train existing state-of-the-art models and obtain a significant performance improvement, of up to 21% and 27%, on the Visual Genome and GQA benchmark datasets, respectively. Furthermore, we showcase the learning efficiency of the proposed framework by demonstrating superior performance in the zero- and few-shot settings where data is scarce.

cs.CV↗

arXiv2020

From Real to Synthetic and Back: Synthesizing Training Data for Multi-Person Scene Understanding

We present a method for synthesizing naturally looking images of multiple people interacting in a specific scenario. These images benefit from the advantages of synthetic data: being fully controllable and fully annotated with any type of standard or custom-defined ground truth. To reduce the synthetic-to-real domain gap, we introduce a pipeline consisting of the following steps: 1) we render scenes in a context modeled after the real world, 2) we train a human parsing model on the synthetic images, 3) we use the model to estimate segmentation maps for real images, 4) we train a conditional generative adversarial network (cGAN) to learn the inverse mapping -- from a segmentation map to a real image, and 5) given new synthetic segmentation maps, we use the cGAN to generate realistic images. An illustration of our pipeline is presented in Figure 2. We use the generated data to train a multi-task model on the challenging tasks of UV mapping and dense depth estimation. We demonstrate the value of the data generation and the trained model, both quantitatively and qualitatively on the CMU Panoptic Dataset.

cs.CV↗