Andrew D. Bagdanov

Xialei Liu, Yu-Song Hu, Xu-Sheng Cao et al.

In class incremental learning (CIL) a model must learn new classes in a sequential manner without forgetting old ones. However, conventional CIL methods consider a balanced distribution for each new task, which ignores the prevalence of long-tailed distributions in the real world. In this work we propose two long-tailed CIL scenarios, which we term ordered and shuffled LT-CIL. Ordered LT-CIL considers the scenario where we learn from head classes collected with more samples than tail classes which have few. Shuffled LT-CIL, on the other hand, assumes a completely random long-tailed distribution for each task. We systematically evaluate existing methods in both LT-CIL scenarios and demonstrate very different behaviors compared to conventional CIL scenarios. Additionally, we propose a two-stage learning baseline with a learnable weight scaling layer for reducing the bias caused by long-tailed distribution in LT-CIL and which in turn also improves the performance of conventional CIL due to the limited exemplars. Our results demonstrate the superior performance (up to 6.44 points in average incremental accuracy) of our approach on CIFAR-100 and ImageNet-Subset. The code is available at https://github.com/xialeiliu/Long-Tailed-CIL

Marco Mistretta, Alberto Baldrati, Marco Bertini et al.

Vision-Language Models (VLMs) demonstrate remarkable zero-shot generalization to unseen tasks, but fall short of the performance of supervised methods in generalizing to downstream tasks with limited data. Prompt learning is emerging as a parameter-efficient method for adapting VLMs, but state-of-the-art approaches require annotated samples. In this paper we propose a novel approach to prompt learning based on unsupervised knowledge distillation from more powerful models. Our approach, which we call Knowledge Distillation Prompt Learning (KDPL), can be integrated into existing prompt learning techniques and eliminates the need for labeled examples during adaptation. Our experiments on more than ten standard benchmark datasets demonstrate that KDPL is very effective at improving generalization of learned prompts for zero-shot domain generalization, zero-shot cross-dataset generalization, and zero-shot base-to-novel class generalization problems. KDPL requires no ground-truth labels for adaptation, and moreover we show that even in the absence of any knowledge of training class names it can be used to effectively transfer knowledge. The code is publicly available at https://github.com/miccunifi/KDPL.

Alex Gomez-Villa, Dipam Goswami, Kai Wang et al.

Exemplar-free class-incremental learning using a backbone trained from scratch and starting from a small first task presents a significant challenge for continual representation learning. Prototype-based approaches, when continually updated, face the critical issue of semantic drift due to which the old class prototypes drift to different positions in the new feature space. Through an analysis of prototype-based continual learning, we show that forgetting is not due to diminished discriminative power of the feature extractor, and can potentially be corrected by drift compensation. To address this, we propose Learnable Drift Compensation (LDC), which can effectively mitigate drift in any moving backbone, whether supervised or unsupervised. LDC is fast and straightforward to integrate on top of existing continual learning approaches. Furthermore, we showcase how LDC can be applied in combination with self-supervised CL methods, resulting in the first exemplar-free semi-supervised continual learning approach. We achieve state-of-the-art performance in both supervised and semi-supervised settings across multiple datasets. Code is available at \url{https://github.com/alviur/ldc}.

Jiang-Tian Zhai, Xialei Liu, Andrew D. Bagdanov et al.

Class Incremental Learning (CIL) aims to sequentially learn new classes while avoiding catastrophic forgetting of previous knowledge. We propose to use Masked Autoencoders (MAEs) as efficient learners for CIL. MAEs were originally designed to learn useful representations through reconstructive unsupervised learning, and they can be easily integrated with a supervised loss for classification. Moreover, MAEs can reliably reconstruct original input images from randomly selected patches, which we use to store exemplars from past tasks more efficiently for CIL. We also propose a bilateral MAE framework to learn from image-level and embedding-level fusion, which produces better-quality reconstructed images and more stable representations. Our experiments confirm that our approach performs better than the state-of-the-art on CIFAR-100, ImageNet-Subset, and ImageNet-Full. The code is available at https://github.com/scok30/MAE-CIL .

Xialei Liu, Jiang-Tian Zhai, Andrew D. Bagdanov et al.

Exemplar-free Class Incremental Learning (EFCIL) aims to sequentially learn tasks with access only to data from the current one. EFCIL is of interest because it mitigates concerns about privacy and long-term storage of data, while at the same time alleviating the problem of catastrophic forgetting in incremental learning. In this work, we introduce task-adaptive saliency for EFCIL and propose a new framework, which we call Task-Adaptive Saliency Supervision (TASS), for mitigating the negative effects of saliency drift between different tasks. We first apply boundary-guided saliency to maintain task adaptivity and \textit{plasticity} on model attention. Besides, we introduce task-agnostic low-level signals as auxiliary supervision to increase the \textit{stability} of model attention. Finally, we introduce a module for injecting and recovering saliency noise to increase the robustness of saliency preservation. Our experiments demonstrate that our method can better preserve saliency maps across tasks and achieve state-of-the-art results on the CIFAR-100, Tiny-ImageNet, and ImageNet-Subset EFCIL benchmarks. Code is available at \url{https://github.com/scok30/tass}.

Simone Magistri, Dipam Goswami, Marco Mistretta et al.

Vision-Language Models like CLIP are extensively used for inter-modal tasks which involve both visual and text modalities. However, when the individual modality encoders are applied to inherently intra-modal tasks like image-to-image retrieval, their performance suffers from the intra-modal misalignment. In this paper we study intra-modal misalignment in CLIP with a focus on the role of the projectors that map pre-projection image and text embeddings into the shared embedding space. By analyzing the form of the cosine similarity applied to projected features, and its interaction with the contrastive CLIP loss, we show that there is an inter-modal operator responsible for aligning the two modalities during training, and a second, intra-modal operator that only enforces intra-modal normalization but does nothing to promote intra-modal alignment. Via spectral analysis of the inter-modal operator, we identify an approximately isotropic subspace in which the two modalities are well-aligned, as well as anisotropic directions specific to each modality. We demonstrate that this aligned subspace can be directly obtained from the projector weights and that removing the anisotropic directions improves intra-modal alignment. Our experiments on intra-modal retrieval and classification benchmarks show that our training-free method reduces intra-modal misalignment, greatly lowers latency, and outperforms existing approaches across multiple pre-trained CLIP-like models. The code is publicly available at: https://github.com/simomagi/IsoCLIP.

Lorenzo Caselli, Marco Mistretta, Simone Magistri et al.

Generalized Category Discovery (GCD) aims to identify novel categories in unlabeled data while leveraging a small labeled subset of known classes. Training a parametric classifier solely on image features often leads to overfitting to old classes, and recent multimodal approaches improve performance by incorporating textual information. However, they treat modalities independently and incur high computational cost. We propose SpectralGCD, an efficient and effective multimodal approach to GCD that uses CLIP cross-modal image-concept similarities as a unified cross-modal representation. Each image is expressed as a mixture over semantic concepts from a large task-agnostic dictionary, which anchors learning to explicit semantics and reduces reliance on spurious visual cues. To maintain the semantic quality of representations learned by an efficient student, we introduce Spectral Filtering which exploits a cross-modal covariance matrix over the softmaxed similarities measured by a strong teacher model to automatically retain only relevant concepts from the dictionary. Forward and reverse knowledge distillation from the same teacher ensures that the cross-modal representations of the student remain both semantically sufficient and well-aligned. Across six benchmarks, SpectralGCD delivers accuracy comparable to or significantly superior to state-of-the-art methods at a fraction of the computational cost. The code is publicly available at: https://github.com/miccunifi/SpectralGCD.

Vaggelis Dorovatas, Malte Schwerin, Andrew D. Bagdanov et al. · mila

Foundation models have transformed machine learning through large-scale pretraining and increased test-time compute. Despite surpassing human performance in several domains, these models remain fundamentally limited in continuous operation, experience accumulation, and personalization, capabilities that are central to adaptive intelligence. While continual learning research has long targeted these goals, its historical focus on in-weight learning (IWL), i.e., updating a single model's parameters to absorb new knowledge, has rendered catastrophic forgetting a persistent challenge. Our position is that combining the strengths of In-Weight Learning (IWL) and the newly emerged capabilities of In-Context Learning (ICL) through the design of modular memory is the missing piece for continual adaptation at scale. We outline a conceptual framework for modular memory-centric architectures that leverage ICL for rapid adaptation and knowledge accumulation, and IWL for stable updates to model capabilities, charting a practical roadmap toward continually learning agents.

Xialei Liu, Xusheng Cao, Haori Lu et al.

With the advent of large-scale pre-trained models, interest in adapting and exploiting them for continual learning scenarios has grown. In this paper, we propose an approach to exploiting pre-trained vision-language models (e.g. CLIP) that enables further adaptation instead of only using zero-shot learning of new tasks. We augment a pre-trained CLIP model with additional layers after the Image Encoder or before the Text Encoder. We investigate three different strategies: a Linear Adapter, a Self-attention Adapter, each operating on the image embedding, and Prompt Tuning which instead modifies prompts input to the CLIP text encoder. We also propose a method for parameter retention in the adapter layers that uses a measure of parameter importance to better maintain stability and plasticity during incremental learning. Our experiments demonstrate that the simplest solution -- a single Linear Adapter layer with parameter retention -- produces the best results. Experiments on several conventional benchmarks consistently show a significant margin of improvement over the current state-of-the-art.

Alessandro Sestini, Joakim Bergdahl, Konrad Tollmar et al.

In games, as in and many other domains, design validation and testing is a huge challenge as systems are growing in size and manual testing is becoming infeasible. This paper proposes a new approach to automated game validation and testing. Our method leverages a data-driven imitation learning technique, which requires little effort and time and no knowledge of machine learning or programming, that designers can use to efficiently train game testing agents. We investigate the validity of our approach through a user study with industry experts. The survey results show that our method is indeed a valid approach to game validation and that data-driven programming would be a useful aid to reducing effort and increasing quality of modern playtesting. The survey also highlights several open challenges. With the help of the most recent literature, we analyze the identified challenges and propose future research directions suitable for supporting and maximizing the utility of our approach.

Marco Cotogni, Fei Yang, Claudio Cusano et al.

We propose a new method for exemplar-free class incremental training of ViTs. The main challenge of exemplar-free continual learning is maintaining plasticity of the learner without causing catastrophic forgetting of previously learned tasks. This is often achieved via exemplar replay which can help recalibrate previous task classifiers to the feature drift which occurs when learning new tasks. Exemplar replay, however, comes at the cost of retaining samples from previous tasks which for many applications may not be possible. To address the problem of continual ViT training, we first propose gated class-attention to minimize the drift in the final ViT transformer block. This mask-based gating is applied to class-attention mechanism of the last transformer block and strongly regulates the weights crucial for previous tasks. Importantly, gated class-attention does not require the task-ID during inference, which distinguishes it from other parameter isolation methods. Secondly, we propose a new method of feature drift compensation that accommodates feature drift in the backbone when learning new tasks. The combination of gated class-attention and cascaded feature drift compensation allows for plasticity towards new tasks while limiting forgetting of previous ones. Extensive experiments performed on CIFAR-100, Tiny-ImageNet and ImageNet100 demonstrate that our exemplar-free method obtains competitive results when compared to rehearsal based ViT methods.

Dipam Goswami, Simone Magistri, Gido M. van de Ven et al.

Vision-language models (VLMs) like CLIP are trained with the objective of aligning text and image pairs. To improve CLIP-based few-shot image classification, recent works have observed that, along with text embeddings, image embeddings from the training set are an important source of information. In this work we investigate the impact of directly mixing image and text prototypes for few-shot classification and analyze this from a bias-variance perspective. We show that mixing prototypes acts like a shrinkage estimator. Although mixed prototypes improve classification performance, the image prototypes still add some noise in the form of instance-specific background or context information. In order to capture only information from the image space relevant to the given classification task, we propose projecting image prototypes onto the principal directions of the semantic text embedding space to obtain a text-aligned semantic image subspace. These text-aligned image prototypes, when mixed with text embeddings, further improve classification. However, for downstream datasets with poor cross-modal alignment in CLIP, semantic alignment might be suboptimal. We show that the image subspace can still be leveraged by modeling the anisotropy using class covariances. We demonstrate that combining a text-aligned mixed prototype classifier and an image-specific LDA classifier outperforms existing methods across few-shot classification benchmarks.

Tomaso Trinci, Simone Magistri, Roberto Verdecchia et al.

With the ever-growing adoption of AI, its impact on the environment is no longer negligible. Despite the potential that continual learning could have towards Green AI, its environmental sustainability remains relatively uncharted. In this work we aim to gain a systematic understanding of the energy efficiency of continual learning algorithms. To that end, we conducted an extensive set of empirical experiments comparing the energy consumption of recent representation-, prompt-, and exemplar-based continual learning algorithms and two standard baseline (fine tuning and joint training) when used to continually adapt a pre-trained ViT-B/16 foundation model. We performed our experiments on three standard datasets: CIFAR-100, ImageNet-R, and DomainNet. Additionally, we propose a novel metric, the Energy NetScore, which we use measure the algorithm efficiency in terms of energy-accuracy trade-off. Through numerous evaluations varying the number and size of the incremental learning steps, our experiments demonstrate that different types of continual learning algorithms have very different impacts on energy consumption during both training and inference. Although often overlooked in the continual learning literature, we found that the energy consumed during the inference phase is crucial for evaluating the environmental sustainability of continual learning models.

Giacomo Cignoni, Simone Magistri, Andrew D. Bagdanov et al.

This paper explores Online Continual Self-Supervised Learning (OCSSL), a scenario in which models learn from continuous streams of unlabeled, non-stationary data, where methods typically employ replay and fast convergence is a central desideratum. We find that OCSSL requires particular attention to the stability-plasticity trade-off: stable methods (e.g. replay with Reservoir sampling) are able to converge faster compared to plastic ones (e.g. FIFO buffer), but incur in performance drops under certain conditions. We explain this collapse phenomenon with the Latent Rehearsal Decay hypothesis, which attributes it to latent space degradation under excessive stability of replay. We introduce two metrics (Overlap and Deviation) that diagnose latent degradation and correlate with accuracy declines. Building on these insights, we propose SOLAR, which leverages efficient online proxies of Deviation to guide buffer management and incorporates an explicit Overlap loss, allowing SOLAR to adaptively managing plasticity. Experiments demonstrate that SOLAR achieves state-of-the-art performance on OCSSL vision benchmarks, with both high convergence speed and final performance.

Girolamo Macaluso, Alessandro Sestini, Andrew D. Bagdanov

Offline Reinforcement Learning (ORL) is a promising approach to reduce the high sample complexity of traditional Reinforcement Learning (RL) by eliminating the need for continuous environmental interactions. ORL exploits a dataset of pre-collected transitions and thus expands the range of application of RL to tasks in which the excessive environment queries increase training time and decrease efficiency, such as in modern AAA games. This paper introduces OfflineMania a novel environment for ORL research. It is inspired by the iconic TrackMania series and developed using the Unity 3D game engine. The environment simulates a single-agent racing game in which the objective is to complete the track through optimal navigation. We provide a variety of datasets to assess ORL performance. These datasets, created from policies of varying ability and in different sizes, aim to offer a challenging testbed for algorithm development and evaluation. We further establish a set of baselines for a range of Online RL, ORL, and hybrid Offline to Online RL approaches using our environment.

Carlo Romeo, Andrew D. Bagdanov

Offline Reinforcement Learning (ORL) offers a robust solution to training agents in applications where interactions with the environment must be strictly limited due to cost, safety, or lack of accurate simulation environments. Despite its potential to facilitate deployment of artificial agents in the real world, Offline Reinforcement Learning typically requires very many demonstrations annotated with ground-truth rewards. Consequently, state-of-the-art ORL algorithms can be difficult or impossible to apply in data-scarce scenarios. In this paper we propose a simple but effective Reward Model that can estimate the reward signal from a very limited sample of environment transitions annotated with rewards. Once the reward signal is modeled, we use the Reward Model to impute rewards for a large sample of reward-free transitions, thus enabling the application of ORL techniques. We demonstrate the potential of our approach on several D4RL continuous locomotion tasks. Our results show that, using only 1\% of reward-labeled transitions from the original datasets, our learned reward model is able to impute rewards for the remaining 99\% of the transitions, from which performant agents can be learned using Offline Reinforcement Learning.

Daniel Marczak, Simone Magistri, Sebastian Cygert et al.

Model merging integrates the weights of multiple task-specific models into a single multi-task model. Despite recent interest in the problem, a significant performance gap between the combined and single-task models remains. In this paper, we investigate the key characteristics of task matrices -- weight update matrices applied to a pre-trained model -- that enable effective merging. We show that alignment between singular components of task-specific and merged matrices strongly correlates with performance improvement over the pre-trained model. Based on this, we propose an isotropic merging framework that flattens the singular value spectrum of task matrices, enhances alignment, and reduces the performance gap. Additionally, we incorporate both common and task-specific subspaces to further improve alignment and performance. Our proposed approach achieves state-of-the-art performance on vision and language tasks across various sets of tasks and model scales. This work advances the understanding of model merging dynamics, offering an effective methodology to merge models without requiring additional training. Code is available at https://github.com/danielm1405/iso-merging .

Carlo Romeo, Andrew D. Bagdanov

Reinforcement learning for legged locomotion has matured into a stack of multi-component reward functions and physics-engine benchmarks whose morphologies are uniformly derived from real commercial hardware. Game NPCs, however, are bound by stylistic constraints absent from sim-to-real robotics and routinely take the form of creatures with no real-robot counterpart. We introduce ARC-RL, a suite of four MuJoCo continuous-control environments featuring robotic morphologies inspired by the bestiary of ARC Raiders: the 18-DoF tall hexapod Queen, the 12-DoF armoured hexapod Bastion, the 18-DoF compact hexapod Tick, and the 12-DoF quadruped Leaper. All four robots share a unified observation template, action convention, simulation cadence, and a single closed-form multi-component reward function whose only per-morphology variation lives in a small set of weights and parameters. The reward fuses a velocity-tracking tent, a healthy survive bonus, a phase-locked gait-compliance bonus/cost pair, action regularisers, three safety penalties, and a posture anchor; no motion-capture data enters the reward at any point. We additionally provide hand-crafted Central Pattern Generator demonstrators per morphology, which serve both as fixed expert references and as sources of prior data for offline-to-online training. On this playground, we conduct a controlled empirical study comparing standard online algorithms (SAC, SPEQ, SOPE-EO) and methods augmented with prior data (SACfD, SPEQ-O2O, SOPE), and characterise how each paradigm copes with the playground's morphological diversity and animation-style stylistic constraints.

Marco Mistretta, Alberto Baldrati, Lorenzo Agnolucci et al.

Pre-trained multi-modal Vision-Language Models like CLIP are widely used off-the-shelf for a variety of applications. In this paper, we show that the common practice of individually exploiting the text or image encoders of these powerful multi-modal models is highly suboptimal for intra-modal tasks like image-to-image retrieval. We argue that this is inherently due to the CLIP-style inter-modal contrastive loss that does not enforce any intra-modal constraints, leading to what we call intra-modal misalignment. To demonstrate this, we leverage two optimization-based modality inversion techniques that map representations from their input modality to the complementary one without any need for auxiliary data or additional trained adapters. We empirically show that, in the intra-modal tasks of image-to-image and text-to-text retrieval, approaching these tasks inter-modally significantly improves performance with respect to intra-modal baselines on more than fifteen datasets. Additionally, we demonstrate that approaching a native inter-modal task (e.g. zero-shot image classification) intra-modally decreases performance, further validating our findings. Finally, we show that incorporating an intra-modal term in the pre-training objective or narrowing the modality gap between the text and image feature embedding spaces helps reduce the intra-modal misalignment. The code is publicly available at: https://github.com/miccunifi/Cross-the-Gap.

Aniello Panariello, Daniel Marczak, Simone Magistri et al.

In this paper, we address the challenges associated with merging low-rank adaptations of large neural networks. With the rise of parameter-efficient adaptation techniques, such as Low-Rank Adaptation (LoRA), model fine-tuning has become more accessible. While fine-tuning models with LoRA is highly efficient, existing merging methods often sacrifice this efficiency by merging fully-sized weight matrices. We propose the Core Space merging framework, which enables the merging of LoRA-adapted models within a common alignment basis, thereby preserving the efficiency of low-rank adaptation while substantially improving accuracy across tasks. We further provide a formal proof that projection into Core Space ensures no loss of information and provide a complexity analysis showing the efficiency gains. Extensive empirical results demonstrate that Core Space significantly improves existing merging techniques and achieves state-of-the-art results on both vision and language tasks while utilizing a fraction of the computational resources. Codebase is available at https://github.com/apanariello4/core-space-merging.

Dipam Goswami, Simone Magistri, Kai Wang et al.

Using pre-trained models has been found to reduce the effect of data heterogeneity and speed up federated learning algorithms. Recent works have explored training-free methods using first- and second-order statistics to aggregate local client data distributions at the server and achieve high performance without any training. In this work, we propose a training-free method based on an unbiased estimator of class covariance matrices which only uses first-order statistics in the form of class means communicated by clients to the server. We show how these estimated class covariances can be used to initialize the global classifier, thus exploiting the covariances without actually sharing them. We also show that using only within-class covariances results in a better classifier initialization. Our approach improves performance in the range of 4-26% with exactly the same communication cost when compared to methods sharing only class means and achieves performance competitive or superior to methods sharing second-order statistics with dramatically less communication overhead. The proposed method is much more communication-efficient than federated prompt-tuning methods and still outperforms them. Finally, using our method to initialize classifiers and then performing federated fine-tuning or linear probing again yields better performance. Code is available at https://github.com/dipamgoswami/FedCOF.

Xialei Liu, Chenshen Wu, Mikel Menta et al.

Humans are capable of learning new tasks without forgetting previous ones, while neural networks fail due to catastrophic forgetting between new and previously-learned tasks. We consider a class-incremental setting which means that the task-ID is unknown at inference time. The imbalance between old and new classes typically results in a bias of the network towards the newest ones. This imbalance problem can either be addressed by storing exemplars from previous tasks, or by using image replay methods. However, the latter can only be applied to toy datasets since image generation for complex datasets is a hard problem. We propose a solution to the imbalance problem based on generative feature replay which does not require any exemplars. To do this, we split the network into two parts: a feature extractor and a classifier. To prevent forgetting, we combine generative feature replay in the classifier with feature distillation in the feature extractor. Through feature generation, our method reduces the complexity of generative replay and prevents the imbalance problem. Our approach is computationally efficient and scalable to large datasets. Experiments confirm that our approach achieves state-of-the-art results on CIFAR-100 and ImageNet, while requiring only a fraction of the storage needed for exemplar-based continual learning. Code available at \url{https://github.com/xialeiliu/GFR-IL}.

Carlo Romeo, Girolamo Macaluso, Alessandro Sestini et al.

Incorporating prior data into online reinforcement learning accelerates training but typically forces a difficult trade-off between high computational costs and long, multi-stage training pipelines. While fixed-length stabilization phases are significantly more computationally efficient than static update schedules, they require task-dependent manual tuning, risking either the waste of prior knowledge or severe overfitting. To address this, we propose SOPE, an algorithm that uses an actor-aligned Off-Policy Policy Evaluation (OPE) signal as an automated early-stopping mechanism to dynamically control the length of offline training phases. By evaluating the critic on a held-out validation split under the current policy's action distribution, SOPE halts gradient updates exactly when out-of-distribution benefits saturate, eliminating the need for manual schedule tuning. Evaluated on 25 continuous control tasks from the Minari benchmark suite, SOPE improves baseline performance by up to 45.6% while reducing the required TFLOPs by up to 22x, thus balancing the tradeoff between sample and computational efficiency. These findings demonstrate that adaptive, evaluation-driven update schedules are more effective than relying on static, exhaustive update schedules.

Simone Magistri, Tomaso Trinci, Albin Soutif-Cormerais et al.

Exemplar-Free Class Incremental Learning (EFCIL) aims to learn from a sequence of tasks without having access to previous task data. In this paper, we consider the challenging Cold Start scenario in which insufficient data is available in the first task to learn a high-quality backbone. This is especially challenging for EFCIL since it requires high plasticity, which results in feature drift which is difficult to compensate for in the exemplar-free setting. To address this problem, we propose a simple and effective approach that consolidates feature representations by regularizing drift in directions highly relevant to previous tasks and employs prototypes to reduce task-recency bias. Our method, called Elastic Feature Consolidation (EFC), exploits a tractable second-order approximation of feature drift based on an Empirical Feature Matrix (EFM). The EFM induces a pseudo-metric in feature space which we use to regularize feature drift in important directions and to update Gaussian prototypes used in a novel asymmetric cross entropy loss which effectively balances prototype rehearsal with data from new tasks. Experimental results on CIFAR-100, Tiny-ImageNet, ImageNet-Subset and ImageNet-1K demonstrate that Elastic Feature Consolidation is better able to learn new tasks by maintaining model plasticity and significantly outperform the state-of-the-art.

Girolamo Macaluso, Alessandro Sestini, Andrew D. Bagdanov

Offline reinforcement learning leverages pre-collected datasets of transitions to train policies. It can serve as effective initialization for online algorithms, enhancing sample efficiency and speeding up convergence. However, when such datasets are limited in size and quality, offline pre-training can produce sub-optimal policies and lead to degraded online reinforcement learning performance. In this paper we propose a model-based data augmentation strategy to maximize the benefits of offline reinforcement learning pre-training and reduce the scale of data needed to be effective. Our approach leverages a world model of the environment trained on the offline dataset to augment states during offline pre-training. We evaluate our approach on a variety of MuJoCo robotic tasks and our results show it can jump-start online fine-tuning and substantially reduce - in some cases by an order of magnitude - the required number of environment interactions.

Carlo Romeo, Girolamo Macaluso, Alessandro Sestini et al.

High update-to-data (UTD) ratio algorithms in reinforcement learning (RL) improve sample efficiency but incur high computational costs, limiting real-world scalability. We propose Offline Stabilization Phases for Efficient Q-Learning (SPEQ), an RL algorithm that combines low-UTD online training with periodic offline stabilization phases. During these phases, Q-functions are fine-tuned with high UTD ratios on a fixed replay buffer, reducing redundant updates on suboptimal data. This structured training schedule optimally balances computational and sample efficiency, addressing the limitations of both high and low UTD ratio approaches. We empirically demonstrate that SPEQ requires from 40% to 99% fewer gradient updates and 27% to 78% less training time compared to state-of-the-art high UTD ratio methods while maintaining or surpassing their performance on the MuJoCo continuous control benchmark. Our findings highlight the potential of periodic stabilization phases as an effective alternative to conventional training schedules, paving the way for more scalable reinforcement learning solutions in real-world applications where computational resources are constrained.

Marco Mistretta, Andrew D. Bagdanov

In this paper we introduce RE-tune, a novel approach for fine-tuning pre-trained Multimodal Biomedical Vision-Language models (VLMs) in Incremental Learning scenarios for multi-label chest disease diagnosis. RE-tune freezes the backbones and only trains simple adaptors on top of the Image and Text encoders of the VLM. By engineering positive and negative text prompts for diseases, we leverage the ability of Large Language Models to steer the training trajectory. We evaluate RE-tune in three realistic incremental learning scenarios: class-incremental, label-incremental, and data-incremental. Our results demonstrate that Biomedical VLMs are natural continual learners and prevent catastrophic forgetting. RE-tune not only achieves accurate multi-label classification results, but also prioritizes patient privacy and it distinguishes itself through exceptional computational efficiency, rendering it highly suitable for broad adoption in real-world healthcare settings.

Samuele Dell'Erba, Andrew D. Bagdanov

Diffusion models have established the state-of-the-art in text-to-image generation, but their performance often relies on a diffusion prior network to translate text embeddings into the visual manifold for easier decoding. These priors are computationally expensive and require extensive training on massive datasets. In this work, we challenge the necessity of a trained prior at all by employing Optimization-based Visual Inversion (OVI), a training-free and zero-shot alternative, to replace the need for a prior. OVI initializes a latent visual representation from random pseudo-tokens and iteratively optimizes it to maximize the cosine similarity with the input textual prompt embedding. We further propose two novel constraints, a Mahalanobis-based and a Nearest-Neighbor loss, to regularize the OVI optimization process toward the distribution of realistic images. Our experiments, conducted on Kandinsky 2.2, show that OVI can serve as an alternative to traditional priors. More importantly, our analysis reveals a critical flaw in current evaluation benchmarks like T2I-CompBench++, where simply using the text embedding as a prior achieves surprisingly high scores, despite lower perceptual quality. Our constrained OVI methods improve visual fidelity over this baseline, with the Nearest-Neighbor approach proving particularly effective. It achieves quantitative scores comparable to or higher than the state-of-the-art data-efficient prior, underscoring the potential of optimization-based strategies as viable, training-free alternatives to traditional priors. The code will be publicly available upon acceptance.

Girolamo Macaluso, Lorenzo Mandelli, Mirko Bicchierai et al.

Diffusion models have recently advanced human motion generation, producing realistic and diverse animations from textual prompts. However, adapting these models to unseen actions or styles typically requires additional motion capture data and full retraining, which is costly and difficult to scale. We propose a post-training framework based on Reinforcement Learning that fine-tunes pretrained motion diffusion models using only textual prompts, without requiring any motion ground truth. Our approach employs a pretrained text-motion retrieval network as a reward signal and optimizes the diffusion policy with Denoising Diffusion Policy Optimization, effectively shifting the model's generative distribution toward the target domain without relying on paired motion data. We evaluate our method on cross-dataset adaptation and leave-one-out motion experiments using the HumanML3D and KIT-ML datasets across both latent- and joint-space diffusion architectures. Results from quantitative metrics and user studies show that our approach consistently improves the quality and diversity of generated motions, while preserving performance on the original distribution. Our approach is a flexible, data-efficient, and privacy-preserving solution for motion adaptation.

Alessandro Sestini, Joakim Bergdahl, Konrad Tollmar et al.

In offline reinforcement learning, agents are trained using only a fixed set of stored transitions derived from a source policy. However, this requires that the dataset be labeled by a reward function. In applied settings such as video game development, the availability of the reward function is not always guaranteed. This paper proposes Trajectory-Ranked OFfline Inverse reinforcement learning (TROFI), a novel approach to effectively learn a policy offline without a pre-defined reward function. TROFI first learns a reward function from human preferences, which it then uses to label the original dataset making it usable for training the policy. In contrast to other approaches, our method does not require optimal trajectories. Through experiments on the D4RL benchmark we demonstrate that TROFI consistently outperforms baselines and performs comparably to using the ground truth reward to learn policies. Additionally, we validate the efficacy of our method in a 3D game environment. Our studies of the reward model highlight the importance of the reward function in this setting: we show that to ensure the alignment of a value function to the actual future discounted reward, it is fundamental to have a well-engineered and easy-to-learn reward function.

Carlo Romeo, Andrew D. Bagdanov

Balancing combat encounters in Dungeons & Dragons (D&D) is a complex task that requires Dungeon Masters (DM) to manually assess party strength, enemy composition, and dynamic player interactions while avoiding interruption of the narrative flow. In this paper, we propose Encounter Generation via Reinforcement Learning (NTRL), a novel approach that automates Dynamic Difficulty Adjustment (DDA) in D&D via combat encounter design. By framing the problem as a contextual bandit, NTRL generates encounters based on real-time party members attributes. In comparison with classic DM heuristics, NTRL iteratively optimizes encounters to extend combat longevity (+200%), increases damage dealt to party members, reducing post-combat hit points (-16.67%), and raises the number of player deaths while maintaining low total party kills (TPK). The intensification of combat forces players to act wisely and engage in tactical maneuvers, even though the generated encounters guarantee high win rates (70%). Even in comparison with encounters designed by human Dungeon Masters, NTRL demonstrates superior performance by enhancing the strategic depth of combat while increasing difficulty in a manner that preserves overall game fairness.

Simone Magistri, Tomaso Trinci, Albin Soutif-Cormerais et al.

Exemplar-free Class Incremental Learning (EFCIL) aims to learn from a sequence of tasks without having access to previous task data. In this paper, we consider the challenging Cold Start scenario in which insufficient data is available in the first task to learn a high-quality backbone. This is especially challenging for EFCIL since it requires high plasticity, resulting in feature drift which is difficult to compensate for in the exemplar-free setting. To address this problem, we propose an effective approach to consolidate feature representations by regularizing drift in directions highly relevant to previous tasks while employing prototypes to reduce task-recency bias. Our approach, which we call Elastic Feature Consolidation++ (EFC++) exploits a tractable second-order approximation of feature drift based on a proposed Empirical Feature Matrix (EFM). The EFM induces a pseudo-metric in feature space which we use to regularize feature drift in important directions and to update Gaussian prototypes. In addition, we introduce a post-training prototype re-balancing phase that updates classifiers to compensate for feature drift. Experimental results on CIFAR-100, Tiny-ImageNet, ImageNet-Subset, ImageNet-1K and DomainNet demonstrate that EFC++ is better able to learn new tasks by maintaining model plasticity and significantly outperforms the state-of-the-art.

Francesc Net, Marc Folia, Pep Casals et al.

In this paper, we address the challenges of automatic metadata annotation in the domain of Galleries, Libraries, Archives, and Museums (GLAMs) by introducing a novel dataset, EUFCC340K, collected from the Europeana portal. Comprising over 340,000 images, the EUFCC340K dataset is organized across multiple facets: Materials, Object Types, Disciplines, and Subjects, following a hierarchical structure based on the Art & Architecture Thesaurus (AAT). We developed several baseline models, incorporating multiple heads on a ConvNeXT backbone for multi-label image tagging on these facets, and fine-tuning a CLIP model with our image text pairs. Our experiments to evaluate model robustness and generalization capabilities in two different test scenarios demonstrate the utility of the dataset in improving multi-label classification tools that have the potential to alleviate cataloging tasks in the cultural heritage sector.

Alessandro Sestini, Linus Gisslén, Joakim Bergdahl et al.

This paper proposes a novel deep reinforcement learning algorithm to perform automatic analysis and detection of gameplay issues in complex 3D navigation environments. The Curiosity-Conditioned Proximal Trajectories (CCPT) method combines curiosity and imitation learning to train agents to methodically explore in the proximity of known trajectories derived from expert demonstrations. We show how CCPT can explore complex environments, discover gameplay issues and design oversights in the process, and recognize and highlight them directly to game designers. We further demonstrate the effectiveness of the algorithm in a novel 3D navigation environment which reflects the complexity of modern AAA video games. Our results show a higher level of coverage and bug discovery than baselines methods, and it hence can provide a valuable tool for game designers to identify issues in game design automatically.

Simone Zini, Alex Gomez-Villa, Marco Buzzelli et al.

Several recent works on self-supervised learning are trained by mapping different augmentations of the same image to the same feature representation. The data augmentations used are of crucial importance to the quality of learned feature representations. In this paper, we analyze how the color jitter traditionally used in data augmentation negatively impacts the quality of the color features in learned feature representations. To address this problem, we propose a more realistic, physics-based color data augmentation - which we call Planckian Jitter - that creates realistic variations in chromaticity and produces a model robust to illumination changes that can be commonly observed in real life, while maintaining the ability to discriminate image content based on color information. Experiments confirm that such a representation is complementary to the representations learned with the currently-used color jitter augmentation and that a simple concatenation leads to significant performance gains on a wide range of downstream datasets. In addition, we present a color sensitivity analysis that documents the impact of different training methods on model neurons and shows that the performance of the learned features is robust with respect to illuminant variations.

Alex Gomez-Villa, Bartlomiej Twardowski, Lu Yu et al.

Recent self-supervised learning methods are able to learn high-quality image representations and are closing the gap with supervised approaches. However, these methods are unable to acquire new knowledge incrementally -- they are, in fact, mostly used only as a pre-training phase over IID data. In this work we investigate self-supervised methods in continual learning regimes without any replay mechanism. We show that naive functional regularization, also known as feature distillation, leads to lower plasticity and limits continual learning performance. Instead, we propose Projected Functional Regularization in which a separate temporal projection network ensures that the newly learned feature space preserves information of the previous one, while at the same time allowing for the learning of new features. This prevents forgetting while maintaining the plasticity of the learner. Comparison with other incremental learning approaches applied to self-supervision demonstrates that our method obtains competitive performance in different scenarios and on multiple datasets.

Alessandro Sestini, Alexander Kuhnle, Andrew D. Bagdanov

In this article we study the problem of training intelligent agents using Reinforcement Learning for the purpose of game development. Unlike systems built to replace human players and to achieve super-human performance, our agents aim to produce meaningful interactions with the player, and at the same time demonstrate behavioral traits as desired by game designers. We show how to combine distinct behavioral policies to obtain a meaningful "fusion" policy which comprises all these behaviors. To this end, we propose four different policy fusion methods for combining pre-trained policies. We further demonstrate how these methods can be used in combination with Inverse Reinforcement Learning in order to create intelligent agents with specific behavioral styles as chosen by game designers, without having to define many and possibly poorly-designed reward functions. Experiments on two different environments indicate that entropy-weighted policy fusion significantly outperforms all others. We provide several practical examples and use-cases for how these methods are indeed useful for video game production and designers.

My Kieu, Lorenzo Berlincioni, Leonardo Galteri et al.

In this paper we propose a method for improving pedestrian detection in the thermal domain using two stages: first, a generative data augmentation approach is used, then a domain adaptation method using generated data adapts an RGB pedestrian detector. Our model, based on the Least-Squares Generative Adversarial Network, is trained to synthesize realistic thermal versions of input RGB images which are then used to augment the limited amount of labeled thermal pedestrian images available for training. We apply our generative data augmentation strategy in order to adapt a pretrained YOLOv3 pedestrian detector to detection in the thermal-only domain. Experimental results demonstrate the effectiveness of our approach: using less than 50\% of available real thermal training data, and relying on synthesized data generated by our model in the domain adaptation phase, our detector achieves state-of-the-art results on the KAIST Multispectral Pedestrian Detection Benchmark; even if more real thermal data is available adding GAN generated images to the training data results in improved performance, thus showing that these images act as an effective form of data augmentation. To the best of our knowledge, our detector achieves the best single-modality detection results on KAIST with respect to the state-of-the-art.

Alessandro Sestini, Alexander Kuhnle, Andrew D. Bagdanov

Recent advances in Deep Reinforcement Learning (DRL) have largely focused on improving the performance of agents with the aim of replacing humans in known and well-defined environments. The use of these techniques as a game design tool for video game production, where the aim is instead to create Non-Player Character (NPC) behaviors, has received relatively little attention until recently. Turn-based strategy games like Roguelikes, for example, present unique challenges to DRL. In particular, the categorical nature of their complex game state, composed of many entities with different attributes, requires agents able to learn how to compare and prioritize these entities. Moreover, this complexity often leads to agents that overfit to states seen during training and that are unable to generalize in the face of design changes made during development. In this paper we propose two network architectures which, when combined with a \emph{procedural loot generation} system, are able to better handle complex categorical state spaces and to mitigate the need for retraining forced by design decisions. The first is based on a dense embedding of the categorical input space that abstracts the discrete observation model and renders trained agents more able to generalize. The second proposed architecture is more general and is based on a Transformer network able to reason relationally about input and input attributes. Our experimental evaluation demonstrates that new agents have better adaptation capacity with respect to a baseline architecture, making this framework more robust to dynamic gameplay changes during development. Based on the results shown in this paper, we believe that these solutions represent a step forward towards making DRL more accessible to the gaming industry.

Alessandro Sestini, Alexander Kuhnle, Andrew D. Bagdanov

Deep Reinforcement Learning achieves very good results in domains where reward functions can be manually engineered. At the same time, there is growing interest within the community in using games based on Procedurally Content Generation (PCG) as benchmark environments since this type of environment is perfect for studying overfitting and generalization of agents under domain shift. Inverse Reinforcement Learning (IRL) can instead extrapolate reward functions from expert demonstrations, with good results even on high-dimensional problems, however there are no examples of applying these techniques to procedurally-generated environments. This is mostly due to the number of demonstrations needed to find a good reward model. We propose a technique based on Adversarial Inverse Reinforcement Learning which can significantly decrease the need for expert demonstrations in PCG games. Through the use of an environment with a limited set of initial seed levels, plus some modifications to stabilize training, we show that our approach, DE-AIRL, is demonstration-efficient and still able to extrapolate reward functions which generalize to the fully procedural domain. We demonstrate the effectiveness of our technique on two procedural environments, MiniGrid and DeepCrawl, for a variety of tasks.

Alessandro Sestini, Alexander Kuhnle, Andrew D. Bagdanov

In this paper we introduce DeepCrawl, a fully-playable Roguelike prototype for iOS and Android in which all agents are controlled by policy networks trained using Deep Reinforcement Learning (DRL). Our aim is to understand whether recent advances in DRL can be used to develop convincing behavioral models for non-player characters in videogames. We begin with an analysis of requirements that such an AI system should satisfy in order to be practically applicable in video game development, and identify the elements of the DRL model used in the DeepCrawl prototype. The successes and limitations of DeepCrawl are documented through a series of playability tests performed on the final game. We believe that the techniques we propose offer insight into innovative new avenues for the development of behaviors for non-player characters in video games, as they offer the potential to overcome critical issues with

Marc Masana, Xialei Liu, Bartlomiej Twardowski et al.

For future learning systems, incremental learning is desirable because it allows for: efficient resource usage by eliminating the need to retrain from scratch at the arrival of new data; reduced memory usage by preventing or limiting the amount of data required to be stored -- also important when privacy limitations are imposed; and learning that more closely resembles human learning. The main challenge for incremental learning is catastrophic forgetting, which refers to the precipitous drop in performance on previously learned tasks after learning a new one. Incremental learning of deep neural networks has seen explosive growth in recent years. Initial work focused on task-incremental learning, where a task-ID is provided at inference time. Recently, we have seen a shift towards class-incremental learning where the learner must discriminate at inference time between all classes seen in previous tasks without recourse to a task-ID. In this paper, we provide a complete survey of existing class-incremental learning methods for image classification, and in particular, we perform an extensive experimental evaluation on thirteen class-incremental methods. We consider several new experimental scenarios, including a comparison of class-incremental methods on multiple large-scale image classification datasets, an investigation into small and large domain shifts, and a comparison of various network architectures.

Riccardo Del Chiaro, Bartłomiej Twardowski, Andrew D. Bagdanov et al.

Research on continual learning has led to a variety of approaches to mitigating catastrophic forgetting in feed-forward classification networks. Until now surprisingly little attention has been focused on continual learning of recurrent models applied to problems like image captioning. In this paper we take a systematic look at continual learning of LSTM-based models for image captioning. We propose an attention-based approach that explicitly accommodates the transient nature of vocabularies in continual image captioning tasks -- i.e. that task vocabularies are not disjoint. We call our method Recurrent Attention to Transient Tasks (RATT), and also show how to adapt continual learning approaches based on weight egularization and knowledge distillation to recurrent continual learning problems. We apply our approaches to incremental image captioning problem on two new continual learning benchmarks we define using the MS-COCO and Flickr30 datasets. Our results demonstrate that RATT is able to sequentially learn five captioning tasks while incurring no forgetting of previously learned ones.

Pietro Bongini, Federico Becattini, Andrew D. Bagdanov et al.

Technology and the fruition of cultural heritage are becoming increasingly more entwined, especially with the advent of smart audio guides, virtual and augmented reality, and interactive installations. Machine learning and computer vision are important components of this ongoing integration, enabling new interaction modalities between user and museum. Nonetheless, the most frequent way of interacting with paintings and statues still remains taking pictures. Yet images alone can only convey the aesthetics of the artwork, lacking is information which is often required to fully understand and appreciate it. Usually this additional knowledge comes both from the artwork itself (and therefore the image depicting it) and from an external source of knowledge, such as an information sheet. While the former can be inferred by computer vision algorithms, the latter needs more structured data to pair visual content with relevant information. Regardless of its source, this information still must be be effectively transmitted to the user. A popular emerging trend in computer vision is Visual Question Answering (VQA), in which users can interact with a neural network by posing questions in natural language and receiving answers about the visual content. We believe that this will be the evolution of smart audio guides for museum visits and simple image browsing on personal smartphones. This will turn the classic audio guide into a smart personal instructor with which the visitor can interact by asking for explanations focused on specific interests. The advantages are twofold: on the one hand the cognitive burden of the visitor will decrease, limiting the flow of information to what the user actually wants to hear; and on the other hand it proposes the most natural way of interacting with a guide, favoring engagement.

Xialei Liu, Joost van de Weijer, Andrew D. Bagdanov

For many applications the collection of labeled data is expensive laborious. Exploitation of unlabeled data during training is thus a long pursued objective of machine learning. Self-supervised learning addresses this by positing an auxiliary task (different, but related to the supervised task) for which data is abundantly available. In this paper, we show how ranking can be used as a proxy task for some regression problems. As another contribution, we propose an efficient backpropagation technique for Siamese networks which prevents the redundant computation introduced by the multi-branch network architecture. We apply our framework to two regression problems: Image Quality Assessment (IQA) and Crowd Counting. For both we show how to automatically generate ranked image sets from unlabeled data. Our results show that networks trained to regress to the ground truth targets for labeled data and to simultaneously learn to rank unlabeled data obtain significantly better, state-of-the-art results for both IQA and crowd counting. In addition, we show that measuring network uncertainty on the self-supervised proxy task is a good measure of informativeness of unlabeled data. This can be used to drive an algorithm for active learning and we show that this reduces labeling effort by up to 50%.

Dena Bazazian, Dimosthenis Karatzas, Andrew D. Bagdanov

Word spotting in natural scene images has many applications in scene understanding and visual assistance. In this paper we propose a technique to create and exploit an intermediate representation of images based on text attributes which are character probability maps. Our representation extends the concept of the Pyramidal Histogram Of Characters (PHOC) by exploiting Fully Convolutional Networks to derive a pixel-wise mapping of the character distribution within candidate word regions. We call this representation the Soft-PHOC. Furthermore, we show how to use Soft-PHOC descriptors for word spotting tasks in egocentric camera streams through an efficient text line proposal algorithm. This is based on the Hough Transform over character attribute maps followed by scoring using Dynamic Time Warping (DTW). We evaluate our results on ICDAR 2015 Challenge 4 dataset of incidental scene text captured by an egocentric camera.

Xialei Liu, Joost van de Weijer, Andrew D. Bagdanov

We propose a novel crowd counting approach that leverages abundantly available unlabeled crowd imagery in a learning-to-rank framework. To induce a ranking of cropped images , we use the observation that any sub-image of a crowded scene image is guaranteed to contain the same number or fewer persons than the super-image. This allows us to address the problem of limited size of existing datasets for crowd counting. We collect two crowd scene datasets from Google using keyword searches and query-by-example image retrieval, respectively. We demonstrate how to efficiently learn from these unlabeled datasets by incorporating learning-to-rank in a multi-task network which simultaneously ranks images and estimates crowd density maps. Experiments on two of the most challenging crowd counting datasets show that our approach obtains state-of-the-art results.

Xialei Liu, Marc Masana, Luis Herranz et al.

In this paper we propose an approach to avoiding catastrophic forgetting in sequential task learning scenarios. Our technique is based on a network reparameterization that approximately diagonalizes the Fisher Information Matrix of the network parameters. This reparameterization takes the form of a factorized rotation of parameter space which, when used in conjunction with Elastic Weight Consolidation (which assumes a diagonal Fisher Information Matrix), leads to significantly better performance on lifelong learning of sequential tasks. Experimental results on the MNIST, CIFAR-100, CUB-200 and Stanford-40 datasets demonstrate that we significantly improve the results of standard elastic weight consolidation, and that we obtain competitive results when compared to other state-of-the-art in lifelong learning without forgetting.

Marc Masana, Joost van de Weijer, Luis Herranz et al.

Deep Neural Networks trained on large datasets can be easily transferred to new domains with far fewer labeled examples by a process called fine-tuning. This has the advantage that representations learned in the large source domain can be exploited on smaller target domains. However, networks designed to be optimal for the source task are often prohibitively large for the target task. In this work we address the compression of networks after domain transfer. We focus on compression algorithms based on low-rank matrix decomposition. Existing methods base compression solely on learned network weights and ignore the statistics of network activations. We show that domain transfer leads to large shifts in network activations and that it is desirable to take this into account when compressing. We demonstrate that considering activation statistics when compressing weights leads to a rank-constrained regression problem with a closed-form solution. Because our method takes into account the target domain, it can more optimally remove the redundancy in the weights. Experiments show that our Domain Adaptive Low Rank (DALR) method significantly outperforms existing low-rank compression techniques. With our approach, the fc6 layer of VGG19 can be compressed more than 4x more than using truncated SVD alone -- with only a minor or no loss in accuracy. When applied to domain-transferred networks it allows for compression down to only 5-20% of the original number of parameters with only a minor drop in performance.

Laura Lopez-Fuentes, Joost van de Weijer, Manuel Gonzalez-Hidalgo et al.

In emergency situations, actions that save lives and limit the impact of hazards are crucial. In order to act, situational awareness is needed to decide what to do. Geolocalized photos and video of the situations as they evolve can be crucial in better understanding them and making decisions faster. Cameras are almost everywhere these days, either in terms of smartphones, installed CCTV cameras, UAVs or others. However, this poses challenges in big data and information overflow. Moreover, most of the time there are no disasters at any given location, so humans aiming to detect sudden situations may not be as alert as needed at any point in time. Consequently, computer vision tools can be an excellent decision support. The number of emergencies where computer vision tools has been considered or used is very wide, and there is a great overlap across related emergency research. Researchers tend to focus on state-of-the-art systems that cover the same emergency as they are studying, obviating important research in other fields. In order to unveil this overlap, the survey is divided along four main axes: the types of emergencies that have been studied in computer vision, the objective that the algorithms can address, the type of hardware needed and the algorithms used. Therefore, this review provides a broad overview of the progress of computer vision covering all sorts of emergencies.