Oron Anschel, Alon Shoshan, Adam Botach, Shunit Haviv Hakimi, Asaf Gendler, Emanuel Ben Baruch, Nadav Bhonker, Igor Kviatkovsky, Manoj Aggarwal, Gerard Medioni. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

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.

With the increasing size of datasets used for training neural networks, data pruning has gained traction in recent years. However, most current data pruning algorithms are limited in their ability to preserve accuracy compared to models trained on the full data, especially in high pruning regimes. In this paper we explore the application of data pruning while incorporating knowledge distillation (KD) when training on a pruned subset. That is, rather than relying solely on ground-truth labels, we also use the soft predictions from a teacher network pre-trained on the complete data. By integrating KD into training, we demonstrate significant improvement across datasets, pruning methods, and on all pruning fractions. We first establish a theoretical motivation for employing self-distillation to improve training on pruned data. Then, we empirically make a compelling and highly practical observation: using KD, simple random pruning is comparable or superior to sophisticated pruning methods across all pruning regimes. On ImageNet for example, we achieve superior accuracy despite training on a random subset of only 50% of the data. Additionally, we demonstrate a crucial connection between the pruning factor and the optimal knowledge distillation weight. This helps mitigate the impact of samples with noisy labels and low-quality images retained by typical pruning algorithms. Finally, we make an intriguing observation: when using lower pruning fractions, larger teachers lead to accuracy degradation, while surprisingly, employing teachers with a smaller capacity than the student’s may improve results. Our code will be made available.

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.

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.

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.

We present a framework for training GANs with explicit control over generated facial images. We are able to control the generated image by settings exact attributes such as age, pose, expression, etc. Most approaches for manipulating 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 (3DMM) to allow fine-grained control capabilities in GANs. Unlike these methods, our control is not constrained to 3DMM 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.

Sparse and redundant representations, where signals are modeled as a combination of a few atoms from an overcomplete dictionary, is increasingly used in many image processing applications, such as denoising, super resolution, and classification. One common problem is learning a "good" dictionary for different tasks. In the classification task the aim is to learn a dictionary that also takes training labels into account, and indeed there exist several approaches to this problem. One well-known technique is D-KSVD, which jointly learns a dictionary and a linear classifier using the K-SVD algorithm. LC-KSVD is a recent variation intended to further improve on this idea by adding an explicit label consistency term to the optimization problem, so that different classes are represented by different dictionary atoms. In this work we prove that, under identical initialization conditions, LC-KSVD with uniform atom allocation is in fact a reformulation of D-KSVD: given the regularization parameters of LC-KSVD, we give a closed-form expression for the equivalent D-KSVD regularization parameter, assuming the LC-KSVD's initialization scheme is used. We confirm this by reproducing several of the original LC-KSVD experiments.

We revisit the problem of specific object recognition using color distributions. In some applications--such as specific person identification--it is highly likely that the color distributions will be multimodal and hence contain a special structure. Although the color distribution changes under different lighting conditions, some aspects of its structure turn out to be invariants. We refer to this structure as an intradistribution structure, and show that it is invariant under a wide range of imaging conditions while being discriminative enough to be practical. Our signature uses shape context descriptors to represent the intradistribution structure. Assuming the widely used diagonal model, we validate that our signature is invariant under certain illumination changes. Experimentally, we use color information as the only cue to obtain good recognition performance on publicly available databases covering both indoor and outdoor conditions. Combining our approach with the complementary covariance descriptor, we demonstrate results exceeding the state-of-the-art performance on the challenging VIPeR and CAVIAR4REID databases.