Mariana-Iuliana Georgescu

Mariana-Iuliana Georgescu, Radu Tudor Ionescu, Andreea-Iuliana Miron et al.

Super-resolving medical images can help physicians in providing more accurate diagnostics. In many situations, computed tomography (CT) or magnetic resonance imaging (MRI) techniques capture several scans (modes) during a single investigation, which can jointly be used (in a multimodal fashion) to further boost the quality of super-resolution results. To this end, we propose a novel multimodal multi-head convolutional attention module to super-resolve CT and MRI scans. Our attention module uses the convolution operation to perform joint spatial-channel attention on multiple concatenated input tensors, where the kernel (receptive field) size controls the reduction rate of the spatial attention, and the number of convolutional filters controls the reduction rate of the channel attention, respectively. We introduce multiple attention heads, each head having a distinct receptive field size corresponding to a particular reduction rate for the spatial attention. We integrate our multimodal multi-head convolutional attention (MMHCA) into two deep neural architectures for super-resolution and conduct experiments on three data sets. Our empirical results show the superiority of our attention module over the state-of-the-art attention mechanisms used in super-resolution. Moreover, we conduct an ablation study to assess the impact of the components involved in our attention module, e.g. the number of inputs or the number of heads. Our code is freely available at https://github.com/lilygeorgescu/MHCA.

Antonio Barbalau, Radu Tudor Ionescu, Mariana-Iuliana Georgescu et al.

A self-supervised multi-task learning (SSMTL) framework for video anomaly detection was recently introduced in literature. Due to its highly accurate results, the method attracted the attention of many researchers. In this work, we revisit the self-supervised multi-task learning framework, proposing several updates to the original method. First, we study various detection methods, e.g. based on detecting high-motion regions using optical flow or background subtraction, since we believe the currently used pre-trained YOLOv3 is suboptimal, e.g. objects in motion or objects from unknown classes are never detected. Second, we modernize the 3D convolutional backbone by introducing multi-head self-attention modules, inspired by the recent success of vision transformers. As such, we alternatively introduce both 2D and 3D convolutional vision transformer (CvT) blocks. Third, in our attempt to further improve the model, we study additional self-supervised learning tasks, such as predicting segmentation maps through knowledge distillation, solving jigsaw puzzles, estimating body pose through knowledge distillation, predicting masked regions (inpainting), and adversarial learning with pseudo-anomalies. We conduct experiments to assess the performance impact of the introduced changes. Upon finding more promising configurations of the framework, dubbed SSMTL++v1 and SSMTL++v2, we extend our preliminary experiments to more data sets, demonstrating that our performance gains are consistent across all data sets. In most cases, our results on Avenue, ShanghaiTech and UBnormal raise the state-of-the-art performance bar to a new level.

Mariana-Iuliana Georgescu, Eduardo Fonseca, Radu Tudor Ionescu et al.

Can we leverage the audiovisual information already present in video to improve self-supervised representation learning? To answer this question, we study various pretraining architectures and objectives within the masked autoencoding framework, motivated by the success of similar methods in natural language and image understanding. We show that we can achieve significant improvements on audiovisual downstream classification tasks, surpassing the state-of-the-art on VGGSound and AudioSet. Furthermore, we can leverage our audiovisual pretraining scheme for multiple unimodal downstream tasks using a single audiovisual pretrained model. We additionally demonstrate the transferability of our representations, achieving state-of-the-art audiovisual results on Epic Kitchens without pretraining specifically for this dataset.

Thomas Hummel, Shyamgopal Karthik, Mariana-Iuliana Georgescu et al.

In Composed Video Retrieval, a video and a textual description which modifies the video content are provided as inputs to the model. The aim is to retrieve the relevant video with the modified content from a database of videos. In this challenging task, the first step is to acquire large-scale training datasets and collect high-quality benchmarks for evaluation. In this work, we introduce EgoCVR, a new evaluation benchmark for fine-grained Composed Video Retrieval using large-scale egocentric video datasets. EgoCVR consists of 2,295 queries that specifically focus on high-quality temporal video understanding. We find that existing Composed Video Retrieval frameworks do not achieve the necessary high-quality temporal video understanding for this task. To address this shortcoming, we adapt a simple training-free method, propose a generic re-ranking framework for Composed Video Retrieval, and demonstrate that this achieves strong results on EgoCVR. Our code and benchmark are freely available at https://github.com/ExplainableML/EgoCVR.

Mariana-Iuliana Georgescu, Radu Tudor Ionescu, Andreea-Iuliana Miron

Medical image segmentation is an actively studied task in medical imaging, where the precision of the annotations is of utter importance towards accurate diagnosis and treatment. In recent years, the task has been approached with various deep learning systems, among the most popular models being U-Net. In this work, we propose a novel strategy to generate ensembles of different architectures for medical image segmentation, by leveraging the diversity (decorrelation) of the models forming the ensemble. More specifically, we utilize the Dice score among model pairs to estimate the correlation between the outputs of the two models forming each pair. To promote diversity, we select models with low Dice scores among each other. We carry out gastro-intestinal tract image segmentation experiments to compare our diversity-promoting ensemble (DiPE) with another strategy to create ensembles based on selecting the top scoring U-Net models. Our empirical results show that DiPE surpasses both individual models as well as the ensemble creation strategy based on selecting the top scoring models.

Mariana-Iuliana Georgescu

Pathological anomalies exhibit diverse appearances in medical imaging, making it difficult to collect and annotate a representative amount of data required to train deep learning models in a supervised setting. Therefore, in this work, we tackle anomaly detection in medical images training our framework using only healthy samples. We propose to use the Masked Autoencoder model to learn the structure of the normal samples, then train an anomaly classifier on top of the difference between the original image and the reconstruction provided by the masked autoencoder. We train the anomaly classifier in a supervised manner using as negative samples the reconstruction of the healthy scans, while as positive samples, we use pseudo-abnormal scans obtained via our novel pseudo-abnormal module. The pseudo-abnormal module alters the reconstruction of the normal samples by changing the intensity of several regions. We conduct experiments on two medical image data sets, namely BRATS2020 and LUNA16 and compare our method with four state-of-the-art anomaly detection frameworks, namely AST, RD4AD, AnoVAEGAN and f-AnoGAN.

Jon Alvarez Justo, Alexandru Ghita, Daniel Kovac et al.

Satellites are increasingly adopting on-board AI to optimize operations and increase autonomy through in-orbit inference. The use of Deep Learning (DL) models for segmentation in hyperspectral imagery offers advantages for remote sensing applications. In this work, we train and test 20 models for multi-class segmentation in hyperspectral imagery, selected for their potential in future space deployment. These models include 1D and 2D Convolutional Neural Networks (CNNs) and the latest vision transformers (ViTs). We propose a lightweight 1D-CNN model, 1D-Justo-LiuNet, which outperforms state-of-the-art models in the hypespectral domain. 1D-Justo-LiuNet exceeds the performance of 2D-CNN UNets and outperforms Apple's lightweight vision transformers designed for mobile inference. 1D-Justo-LiuNet achieves the highest accuracy (0.93) with the smallest model size (4,563 parameters) among all tested models, while maintaining fast inference. Unlike 2D-CNNs and ViTs, which encode both spectral and spatial information, 1D-Justo-LiuNet focuses solely on the rich spectral features in hyperspectral data, benefitting from the high-dimensional feature space. Our findings are validated across various satellite datasets, with the HYPSO-1 mission serving as the primary case study for sea, land, and cloud segmentation. We further confirm our conclusions through generalization tests on other hyperspectral missions, such as NASA's EO-1. Based on its superior performance and compact size, we conclude that 1D-Justo-LiuNet is highly suitable for in-orbit deployment, providing an effective solution for optimizing and automating satellite operations at edge.

Rafael-Edy Menadil, Mariana-Iuliana Georgescu, Radu Tudor Ionescu

Learning Using Privileged Information is a particular type of knowledge distillation where the teacher model benefits from an additional data representation during training, called privileged information, improving the student model, which does not see the extra representation. However, privileged information is rarely available in practice. To this end, we propose a text classification framework that harnesses text-to-image diffusion models to generate artificial privileged information. The generated images and the original text samples are further used to train multimodal teacher models based on state-of-the-art transformer-based architectures. Finally, the knowledge from multimodal teachers is distilled into a text-based (unimodal) student. Hence, by employing a generative model to produce synthetic data as privileged information, we guide the training of the student model. Our framework, called Learning Using Generated Privileged Information (LUGPI), yields noticeable performance gains on four text classification data sets, demonstrating its potential in text classification without any additional cost during inference.

Vlad-Constantin Lungu-Stan, Ionut Mironica, Mariana-Iuliana Georgescu

Media design layer generation enables the creation of fully editable, layered design documents such as posters, flyers, and logos using only natural language prompts. Existing methods either restrict outputs to a fixed number of layers or require each layer to contain only spatially continuous regions, causing the layer count to scale linearly with design complexity. We propose LaDe (Layered Media Design), a latent diffusion framework that generates a flexible number of semantically meaningful layers. LaDe combines three components: an LLM-based prompt expander that transforms a short user intent into structured per-layer descriptions that guide the generation, a Latent Diffusion Transformer with a 4D RoPE positional encoding mechanism that jointly generates the full media design and its constituent RGBA layers, and an RGBA VAE that decodes each layer with full alpha-channel support. By conditioning on layer samples during training, our unified framework supports three tasks: text-to-image generation, text-to-layers media design generation, and media design decomposition. We compare LaDe to Qwen-Image-Layered on text-to-layers and image-to-layers tasks on the Crello test set. LaDe outperforms Qwen-Image-Layered in text-to-layers generation by improving text-to-layer alignment, as validated by two VLM-as-a-judge evaluators (GPT-4o mini and Qwen3-VL).

Yuqian Zheng, Mariana-Iuliana Georgescu

Composed Video Retrieval (CoVR) facilitates video retrieval by combining visual and textual queries. However, existing CoVR frameworks typically fuse multimodal inputs in a single stage, achieving only marginal gains over initial baseline. To address this, we propose a novel CoVR framework that leverages the representational power of Vision Language Models (VLMs). Our framework incorporates a novel cross-attention module X-Aligner, composed of cross-attention layers that progressively fuse visual and textual inputs and align their multimodal representation with that of the target video. To further enhance the representation of the multimodal query, we incorporate the caption of the visual query as an additional input. The framework is trained in two stages to preserve the pretrained VLM representation. In the first stage, only the newly introduced module is trained, while in the second stage, the textual query encoder is also fine-tuned. We implement our framework on top of BLIP-family architecture, namely BLIP and BLIP-2, and train it on the Webvid-CoVR data set. In addition to in-domain evaluation on Webvid-CoVR-Test, we perform zero-shot evaluations on the Composed Image Retrieval (CIR) data sets CIRCO and Fashion-IQ. Our framework achieves state-of-the-art performance on CoVR obtaining a Recall@1 of 63.93% on Webvid-CoVR-Test, and demonstrates strong zero-shot generalization on CIR tasks.

Rui Xiao, Sanghwan Kim, Mariana-Iuliana Georgescu et al.

CLIP has shown impressive results in aligning images and texts at scale. However, its ability to capture detailed visual features remains limited because CLIP matches images and texts at a global level. To address this issue, we propose FLAIR, Fine-grained Language-informed Image Representations, an approach that utilizes long and detailed image descriptions to learn localized image embeddings. By sampling diverse sub-captions that describe fine-grained details about an image, we train our vision-language model to produce not only global embeddings but also text-specific image representations. Our model introduces text-conditioned attention pooling on top of local image tokens to produce fine-grained image representations that excel at retrieving detailed image content. We achieve state-of-the-art performance on both, existing multimodal retrieval benchmarks, as well as, our newly introduced fine-grained retrieval task which evaluates vision-language models' ability to retrieve partial image content. Furthermore, our experiments demonstrate the effectiveness of FLAIR trained on 30M image-text pairs in capturing fine-grained visual information, including zero-shot semantic segmentation, outperforming models trained on billions of pairs. Code is available at https://github.com/ExplainableML/flair .

Sanghwan Kim, Rui Xiao, Mariana-Iuliana Georgescu et al.

Vision-Language Models (VLMs) trained with contrastive loss have achieved significant advancements in various vision and language tasks. However, the global nature of the contrastive loss makes VLMs focus predominantly on foreground objects, neglecting other crucial information in the image, which limits their effectiveness in downstream tasks. To address these challenges, we propose COSMOS: CrOSs-MOdality Self-distillation for vision-language pre-training that integrates a novel text-cropping strategy and cross-attention module into a self-supervised learning framework. We create global and local views of images and texts (i.e., multi-modal augmentations), which are essential for self-distillation in VLMs. We further introduce a cross-attention module, enabling COSMOS to learn comprehensive cross-modal representations optimized via a cross-modality self-distillation loss. COSMOS consistently outperforms previous strong baselines on various zero-shot downstream tasks, including retrieval, classification, and semantic segmentation. Additionally, it surpasses CLIP-based models trained on larger datasets in visual perception and contextual understanding tasks. Code is available at https://github.com/ExplainableML/cosmos.

Diana-Nicoleta Grigore, Mariana-Iuliana Georgescu, Jon Alvarez Justo et al.

Few-shot knowledge distillation recently emerged as a viable approach to harness the knowledge of large-scale pre-trained models, using limited data and computational resources. In this paper, we propose a novel few-shot feature distillation approach for vision transformers. Our approach is based on two key steps. Leveraging the fact that vision transformers have a consistent depth-wise structure, we first copy the weights from intermittent layers of existing pre-trained vision transformers (teachers) into shallower architectures (students), where the intermittence factor controls the complexity of the student transformer with respect to its teacher. Next, we employ an enhanced version of Low-Rank Adaptation (LoRA) to distill knowledge into the student in a few-shot scenario, aiming to recover the information processing carried out by the skipped teacher layers. We present comprehensive experiments with supervised and self-supervised transformers as teachers, on six data sets from various domains (natural, medical and satellite images) and tasks (classification and segmentation). The empirical results confirm the superiority of our approach over state-of-the-art competitors. Moreover, the ablation results demonstrate the usefulness of each component of the proposed pipeline. We release our code at https://github.com/dianagrigore/WeCoLoRA.

Francesco Dalmonte, Emirhan Bayar, Emre Akbas et al.

Anomaly detection in medical images is an important yet challenging task due to the diversity of possible anomalies and the practical impossibility of collecting comprehensively annotated data sets. In this work, we tackle unsupervised medical anomaly detection proposing a modernized autoencoder-based framework, the Q-Former Autoencoder, that leverages state-of-the-art pretrained vision foundation models, such as DINO, DINOv2 and Masked Autoencoder. Instead of training encoders from scratch, we directly utilize frozen vision foundation models as feature extractors, enabling rich, multi-stage, high-level representations without domain-specific fine-tuning. We propose the usage of the Q-Former architecture as the bottleneck, which enables the control of the length of the reconstruction sequence, while efficiently aggregating multiscale features. Additionally, we incorporate a perceptual loss computed using features from a pretrained Masked Autoencoder, guiding the reconstruction towards semantically meaningful structures. Our framework is evaluated on four diverse medical anomaly detection benchmarks, achieving state-of-the-art results on BraTS2021, RESC, and RSNA. Our results highlight the potential of vision foundation model encoders, pretrained on natural images, to generalize effectively to medical image analysis tasks without further fine-tuning. We release the code and models at https://github.com/emirhanbayar/QFAE.

Andra Acsintoae, Andrei Florescu, Mariana-Iuliana Georgescu et al.

Detecting abnormal events in video is commonly framed as a one-class classification task, where training videos contain only normal events, while test videos encompass both normal and abnormal events. In this scenario, anomaly detection is an open-set problem. However, some studies assimilate anomaly detection to action recognition. This is a closed-set scenario that fails to test the capability of systems at detecting new anomaly types. To this end, we propose UBnormal, a new supervised open-set benchmark composed of multiple virtual scenes for video anomaly detection. Unlike existing data sets, we introduce abnormal events annotated at the pixel level at training time, for the first time enabling the use of fully-supervised learning methods for abnormal event detection. To preserve the typical open-set formulation, we make sure to include disjoint sets of anomaly types in our training and test collections of videos. To our knowledge, UBnormal is the first video anomaly detection benchmark to allow a fair head-to-head comparison between one-class open-set models and supervised closed-set models, as shown in our experiments. Moreover, we provide empirical evidence showing that UBnormal can enhance the performance of a state-of-the-art anomaly detection framework on two prominent data sets, Avenue and ShanghaiTech. Our benchmark is freely available at https://github.com/lilygeorgescu/UBnormal.

Nicolae-Catalin Ristea, Andreea-Iuliana Miron, Olivian Savencu et al.

We propose a novel approach to translate unpaired contrast computed tomography (CT) scans to non-contrast CT scans and the other way around. Solving this task has two important applications: (i) to automatically generate contrast CT scans for patients for whom injecting contrast substance is not an option, and (ii) to enhance the alignment between contrast and non-contrast CT by reducing the differences induced by the contrast substance before registration. Our approach is based on cycle-consistent generative adversarial convolutional transformers, for short, CyTran. Our neural model can be trained on unpaired images, due to the integration of a multi-level cycle-consistency loss. Aside from the standard cycle-consistency loss applied at the image level, we propose to apply additional cycle-consistency losses between intermediate feature representations, which enforces the model to be cycle-consistent at multiple representations levels, leading to superior results. To deal with high-resolution images, we design a hybrid architecture based on convolutional and multi-head attention layers. In addition, we introduce a novel data set, Coltea-Lung-CT-100W, containing 100 3D triphasic lung CT scans (with a total of 37,290 images) collected from 100 female patients (there is one examination per patient). Each scan contains three phases (non-contrast, early portal venous, and late arterial), allowing us to perform experiments to compare our novel approach with state-of-the-art methods for image style transfer. Our empirical results show that CyTran outperforms all competing methods. Moreover, we show that CyTran can be employed as a preliminary step to improve a state-of-the-art medical image alignment method. We release our novel model and data set as open source at https://github.com/ristea/cycle-transformer.

Tudor Mare, Georgian Duta, Mariana-Iuliana Georgescu et al.

The COVID-19 pandemic raises the problem of adapting face recognition systems to the new reality, where people may wear surgical masks to cover their noses and mouths. Traditional data sets (e.g., CelebA, CASIA-WebFace) used for training these systems were released before the pandemic, so they now seem unsuited due to the lack of examples of people wearing masks. We propose a method for enhancing data sets containing faces without masks by creating synthetic masks and overlaying them on faces in the original images. Our method relies on SparkAR Studio, a developer program made by Facebook that is used to create Instagram face filters. In our approach, we use 9 masks of different colors, shapes and fabrics. We employ our method to generate a number of 445,446 (90%) samples of masks for the CASIA-WebFace data set and 196,254 (96.8%) masks for the CelebA data set, releasing the mask images at https://github.com/securifai/masked_faces. We show that our method produces significantly more realistic training examples of masks overlaid on faces by asking volunteers to qualitatively compare it to other methods or data sets designed for the same task. We also demonstrate the usefulness of our method by evaluating state-of-the-art face recognition systems (FaceNet, VGG-face, ArcFace) trained on our enhanced data sets and showing that they outperform equivalent systems trained on original data sets (containing faces without masks) or competing data sets (containing masks generated by related methods), when the test benchmarks contain masked faces.

Mariana-Iuliana Georgescu, Radu Tudor Ionescu, Nicolae-Catalin Ristea et al.

In order to classify linearly non-separable data, neurons are typically organized into multi-layer neural networks that are equipped with at least one hidden layer. Inspired by some recent discoveries in neuroscience, we propose a new model of artificial neuron along with a novel activation function enabling the learning of nonlinear decision boundaries using a single neuron. We show that a standard neuron followed by our novel apical dendrite activation (ADA) can learn the XOR logical function with 100% accuracy. Furthermore, we conduct experiments on six benchmark data sets from computer vision, signal processing and natural language processing, i.e. MOROCO, UTKFace, CREMA-D, Fashion-MNIST, Tiny ImageNet and ImageNet, showing that the ADA and the leaky ADA functions provide superior results to Rectified Linear Units (ReLU), leaky ReLU, RBF and Swish, for various neural network architectures, e.g. one-hidden-layer or two-hidden-layer multi-layer perceptrons (MLPs) and convolutional neural networks (CNNs) such as LeNet, VGG, ResNet and Character-level CNN. We obtain further performance improvements when we change the standard model of the neuron with our pyramidal neuron with apical dendrite activations (PyNADA). Our code is available at: https://github.com/raduionescu/pynada.

Ronald Skorobogat, Karsten Roth, Mariana-Iuliana Georgescu

Model merging enables the combination of multiple specialized expert models into a single model capable of performing multiple tasks. However, the benefits of merging an increasing amount of specialized experts generally lead to diminishing returns and reduced overall performance gains. In this work, we offer an explanation and analysis from a task arithmetic perspective; revealing that as the merging process (across numerous existing merging methods) continues for more and more experts, the associated task vector space experiences rank collapse. To mitigate this issue, we introduce Subspace Boosting, which operates on the singular value decomposed task vector space and maintains task vector ranks. Subspace Boosting raises merging efficacy for up to 20 expert models by large margins of more than 10% when evaluated on both vision and language benchmarks. Moreover, we propose employing Higher-Order Generalized Singular Value Decomposition to quantify task similarity, offering a new interpretable perspective on model merging.

Shyam Nandan Rai, Shyamgopal Karthik, Mariana-Iuliana Georgescu et al.

With the growing deployment of autonomous driving agents, the detection and segmentation of road obstacles have become critical to ensure safe autonomous navigation. However, existing road-obstacle segmentation methods are applied on individual frames, overlooking the temporal nature of the problem, leading to inconsistent prediction maps between consecutive frames. In this work, we demonstrate that the road-obstacle segmentation task is inherently temporal, since the segmentation maps for consecutive frames are strongly correlated. To address this, we curate and adapt four evaluation benchmarks for road-obstacle video segmentation and evaluate 11 state-of-the-art image- and video-based segmentation methods on these benchmarks. Moreover, we introduce two strong baseline methods based on vision foundation models. Our approach establishes a new state-of-the-art in road-obstacle video segmentation for long-range video sequences, providing valuable insights and direction for future research.

Adrian Sandru, Mariana-Iuliana Georgescu, Radu Tudor Ionescu

Recent studies revealed that convolutional neural networks do not generalize well to small image transformations, e.g. rotations by a few degrees or translations of a few pixels. To improve the robustness to such transformations, we propose to introduce data augmentation at intermediate layers of the neural architecture, in addition to the common data augmentation applied on the input images. By introducing small perturbations to activation maps (features) at various levels, we develop the capacity of the neural network to cope with such transformations. We conduct experiments on three image classification benchmarks (Tiny ImageNet, Caltech-256 and Food-101), considering two different convolutional architectures (ResNet-18 and DenseNet-121). When compared with two state-of-the-art stabilization methods, the empirical results show that our approach consistently attains the best trade-off between accuracy and mean flip rate.

Mariana-Iuliana Georgescu, Georgian Duta, Radu Tudor Ionescu

We study a series of recognition tasks in two realistic scenarios requiring the analysis of faces under strong occlusion. On the one hand, we aim to recognize facial expressions of people wearing Virtual Reality (VR) headsets. On the other hand, we aim to estimate the age and identify the gender of people wearing surgical masks. For all these tasks, the common ground is that half of the face is occluded. In this challenging setting, we show that convolutional neural networks (CNNs) trained on fully-visible faces exhibit very low performance levels. While fine-tuning the deep learning models on occluded faces is extremely useful, we show that additional performance gains can be obtained by distilling knowledge from models trained on fully-visible faces. To this end, we study two knowledge distillation methods, one based on teacher-student training and one based on triplet loss. Our main contribution consists in a novel approach for knowledge distillation based on triplet loss, which generalizes across models and tasks. Furthermore, we consider combining distilled models learned through conventional teacher-student training or through our novel teacher-student training based on triplet loss. We provide empirical evidence showing that, in most cases, both individual and combined knowledge distillation methods bring statistically significant performance improvements. We conduct experiments with three different neural models (VGG-f, VGG-face, ResNet-50) on various tasks (facial expression recognition, gender recognition, age estimation), showing consistent improvements regardless of the model or task.

Mariana-Iuliana Georgescu, Antonio Barbalau, Radu Tudor Ionescu et al.

Anomaly detection in video is a challenging computer vision problem. Due to the lack of anomalous events at training time, anomaly detection requires the design of learning methods without full supervision. In this paper, we approach anomalous event detection in video through self-supervised and multi-task learning at the object level. We first utilize a pre-trained detector to detect objects. Then, we train a 3D convolutional neural network to produce discriminative anomaly-specific information by jointly learning multiple proxy tasks: three self-supervised and one based on knowledge distillation. The self-supervised tasks are: (i) discrimination of forward/backward moving objects (arrow of time), (ii) discrimination of objects in consecutive/intermittent frames (motion irregularity) and (iii) reconstruction of object-specific appearance information. The knowledge distillation task takes into account both classification and detection information, generating large prediction discrepancies between teacher and student models when anomalies occur. To the best of our knowledge, we are the first to approach anomalous event detection in video as a multi-task learning problem, integrating multiple self-supervised and knowledge distillation proxy tasks in a single architecture. Our lightweight architecture outperforms the state-of-the-art methods on three benchmarks: Avenue, ShanghaiTech and UCSD Ped2. Additionally, we perform an ablation study demonstrating the importance of integrating self-supervised learning and normality-specific distillation in a multi-task learning setting.

Adrian Sandru, Georgian-Emilian Duta, Mariana-Iuliana Georgescu et al.

We propose a deep learning method to automatically detect personal protective equipment (PPE), such as helmets, surgical masks, reflective vests, boots and so on, in images of people. Typical approaches for PPE detection based on deep learning are (i) to train an object detector for items such as those listed above or (ii) to train a person detector and a classifier that takes the bounding boxes predicted by the detector and discriminates between people wearing and people not wearing the corresponding PPE items. We propose a novel and accurate approach that uses three components: a person detector, a body pose estimator and a classifier. Our novelty consists in using the pose estimator only at training time, to improve the prediction performance of the classifier. We modify the neural architecture of the classifier by adding a spatial attention mechanism, which is trained using supervision signal from the pose estimator. In this way, the classifier learns to focus on PPE items, using knowledge from the pose estimator with almost no computational overhead during inference.

Mariana-Iuliana Georgescu, Radu Tudor Ionescu, Fahad Shahbaz Khan et al.

Abnormal event detection in video is a complex computer vision problem that has attracted significant attention in recent years. The complexity of the task arises from the commonly-adopted definition of an abnormal event, that is, a rarely occurring event that typically depends on the surrounding context. Following the standard formulation of abnormal event detection as outlier detection, we propose a background-agnostic framework that learns from training videos containing only normal events. Our framework is composed of an object detector, a set of appearance and motion auto-encoders, and a set of classifiers. Since our framework only looks at object detections, it can be applied to different scenes, provided that normal events are defined identically across scenes and that the single main factor of variation is the background. To overcome the lack of abnormal data during training, we propose an adversarial learning strategy for the auto-encoders. We create a scene-agnostic set of out-of-domain pseudo-abnormal examples, which are correctly reconstructed by the auto-encoders before applying gradient ascent on the pseudo-abnormal examples. We further utilize the pseudo-abnormal examples to serve as abnormal examples when training appearance-based and motion-based binary classifiers to discriminate between normal and abnormal latent features and reconstructions. We compare our framework with the state-of-the-art methods on four benchmark data sets, using various evaluation metrics. Compared to existing methods, the empirical results indicate that our approach achieves favorable performance on all data sets. In addition, we provide region-based and track-based annotations for two large-scale abnormal event detection data sets from the literature, namely ShanghaiTech and Subway.

Mariana-Iuliana Georgescu, Radu Tudor Ionescu

In this paper, we study the task of facial expression recognition under strong occlusion. We are particularly interested in cases where 50% of the face is occluded, e.g. when the subject wears a Virtual Reality (VR) headset. While previous studies show that pre-training convolutional neural networks (CNNs) on fully-visible (non-occluded) faces improves the accuracy, we propose to employ knowledge distillation to achieve further improvements. First of all, we employ the classic teacher-student training strategy, in which the teacher is a CNN trained on fully-visible faces and the student is a CNN trained on occluded faces. Second of all, we propose a new approach for knowledge distillation based on triplet loss. During training, the goal is to reduce the distance between an anchor embedding, produced by a student CNN that takes occluded faces as input, and a positive embedding (from the same class as the anchor), produced by a teacher CNN trained on fully-visible faces, so that it becomes smaller than the distance between the anchor and a negative embedding (from a different class than the anchor), produced by the student CNN. Third of all, we propose to combine the distilled embeddings obtained through the classic teacher-student strategy and our novel teacher-student strategy based on triplet loss into a single embedding vector. We conduct experiments on two benchmarks, FER+ and AffectNet, with two CNN architectures, VGG-f and VGG-face, showing that knowledge distillation can bring significant improvements over the state-of-the-art methods designed for occluded faces in the VR setting.

Mariana-Iuliana Georgescu, Radu Tudor Ionescu, Nicolae Verga

CT scanners that are commonly-used in hospitals nowadays produce low-resolution images, up to 512 pixels in size. One pixel in the image corresponds to a one millimeter piece of tissue. In order to accurately segment tumors and make treatment plans, doctors need CT scans of higher resolution. The same problem appears in MRI. In this paper, we propose an approach for the single-image super-resolution of 3D CT or MRI scans. Our method is based on deep convolutional neural networks (CNNs) composed of 10 convolutional layers and an intermediate upscaling layer that is placed after the first 6 convolutional layers. Our first CNN, which increases the resolution on two axes (width and height), is followed by a second CNN, which increases the resolution on the third axis (depth). Different from other methods, we compute the loss with respect to the ground-truth high-resolution output right after the upscaling layer, in addition to computing the loss after the last convolutional layer. The intermediate loss forces our network to produce a better output, closer to the ground-truth. A widely-used approach to obtain sharp results is to add Gaussian blur using a fixed standard deviation. In order to avoid overfitting to a fixed standard deviation, we apply Gaussian smoothing with various standard deviations, unlike other approaches. We evaluate our method in the context of 2D and 3D super-resolution of CT and MRI scans from two databases, comparing it to relevant related works from the literature and baselines based on various interpolation schemes, using 2x and 4x scaling factors. The empirical results show that our approach attains superior results to all other methods. Moreover, our human annotation study reveals that both doctors and regular annotators chose our method in favor of Lanczos interpolation in 97.55% cases for 2x upscaling factor and in 96.69% cases for 4x upscaling factor.

Mariana-Iuliana Georgescu, Radu Tudor Ionescu

In this paper, we present an approach based on convolutional neural networks (CNNs) for facial expression recognition in a difficult setting with severe occlusions. More specifically, our task is to recognize the facial expression of a person wearing a virtual reality (VR) headset which essentially occludes the upper part of the face. In order to accurately train neural networks for this setting, in which faces are severely occluded, we modify the training examples by intentionally occluding the upper half of the face. This forces the neural networks to focus on the lower part of the face and to obtain better accuracy rates than models trained on the entire faces. Our empirical results on two benchmark data sets, FER+ and AffectNet, show that our CNN models' predictions on lower-half faces are up to 13% higher than the baseline CNN models trained on entire faces, proving their suitability for the VR setting. Furthermore, our models' predictions on lower-half faces are no more than 10% under the baseline models' predictions on full faces, proving that there are enough clues in the lower part of the face to accurately predict facial expressions.

Mariana-Iuliana Georgescu, Radu Tudor Ionescu

We propose a novel agglomerative clustering method based on unmasking, a technique that was previously used for authorship verification of text documents and for abnormal event detection in videos. In order to join two clusters, we alternate between (i) training a binary classifier to distinguish between the samples from one cluster and the samples from the other cluster, and (ii) removing at each step the most discriminant features. The faster-decreasing accuracy rates of the intermediately-obtained classifiers indicate that the two clusters should be joined. To the best of our knowledge, this is the first work to apply unmasking in order to cluster images. We compare our method with k-means as well as a recent state-of-the-art clustering method. The empirical results indicate that our approach is able to improve performance for various (deep and shallow) feature representations and different tasks, such as handwritten digit recognition, texture classification and fine-grained object recognition.

Radu Tudor Ionescu, Fahad Shahbaz Khan, Mariana-Iuliana Georgescu et al.

Abnormal event detection in video is a challenging vision problem. Most existing approaches formulate abnormal event detection as an outlier detection task, due to the scarcity of anomalous data during training. Because of the lack of prior information regarding abnormal events, these methods are not fully-equipped to differentiate between normal and abnormal events. In this work, we formalize abnormal event detection as a one-versus-rest binary classification problem. Our contribution is two-fold. First, we introduce an unsupervised feature learning framework based on object-centric convolutional auto-encoders to encode both motion and appearance information. Second, we propose a supervised classification approach based on clustering the training samples into normality clusters. A one-versus-rest abnormal event classifier is then employed to separate each normality cluster from the rest. For the purpose of training the classifier, the other clusters act as dummy anomalies. During inference, an object is labeled as abnormal if the highest classification score assigned by the one-versus-rest classifiers is negative. Comprehensive experiments are performed on four benchmarks: Avenue, ShanghaiTech, UCSD and UMN. Our approach provides superior results on all four data sets. On the large-scale ShanghaiTech data set, our method provides an absolute gain of 8.4% in terms of frame-level AUC compared to the state-of-the-art method [Sultani et al., CVPR 2018].

Mariana-Iuliana Georgescu, Radu Tudor Ionescu, Marius Popescu

We present an approach that combines automatic features learned by convolutional neural networks (CNN) and handcrafted features computed by the bag-of-visual-words (BOVW) model in order to achieve state-of-the-art results in facial expression recognition. To obtain automatic features, we experiment with multiple CNN architectures, pre-trained models and training procedures, e.g. Dense-Sparse-Dense. After fusing the two types of features, we employ a local learning framework to predict the class label for each test image. The local learning framework is based on three steps. First, a k-nearest neighbors model is applied in order to select the nearest training samples for an input test image. Second, a one-versus-all Support Vector Machines (SVM) classifier is trained on the selected training samples. Finally, the SVM classifier is used to predict the class label only for the test image it was trained for. Although we have used local learning in combination with handcrafted features in our previous work, to the best of our knowledge, local learning has never been employed in combination with deep features. The experiments on the 2013 Facial Expression Recognition (FER) Challenge data set, the FER+ data set and the AffectNet data set demonstrate that our approach achieves state-of-the-art results. With a top accuracy of 75.42% on FER 2013, 87.76% on the FER+, 59.58% on AffectNet 8-way classification and 63.31% on AffectNet 7-way classification, we surpass the state-of-the-art methods by more than 1% on all data sets.