Marco Cannici

Yifei Liu, Mathias Gehrig, Nico Messikommer et al.

Vision Transformers (ViTs) have shown impressive performance in computer vision, but their high computational cost, quadratic in the number of tokens, limits their adoption in computation-constrained applications. However, this large number of tokens may not be necessary, as not all tokens are equally important. In this paper, we investigate token pruning to accelerate inference for object detection and instance segmentation, extending prior works from image classification. Through extensive experiments, we offer four insights for dense tasks: (i) tokens should not be completely pruned and discarded, but rather preserved in the feature maps for later use. (ii) reactivating previously pruned tokens can further enhance model performance. (iii) a dynamic pruning rate based on images is better than a fixed pruning rate. (iv) a lightweight, 2-layer MLP can effectively prune tokens, achieving accuracy comparable with complex gating networks with a simpler design. We assess the effects of these design decisions on the COCO dataset and introduce an approach that incorporates these findings, showing a reduction in performance decline from ~1.5 mAP to ~0.3 mAP in both boxes and masks, compared to existing token pruning methods. In relation to the dense counterpart that utilizes all tokens, our method realizes an increase in inference speed, achieving up to 34% faster performance for the entire network and 46% for the backbone.

Ling Gao, Hang Su, Daniel Gehrig et al.

Event-based cameras are ideal for line-based motion estimation, since they predominantly respond to edges in the scene. However, accurately determining the camera displacement based on events continues to be an open problem. This is because line feature extraction and dynamics estimation are tightly coupled when using event cameras, and no precise model is currently available for describing the complex structures generated by lines in the space-time volume of events. We solve this problem by deriving the correct non-linear parametrization of such manifolds, which we term eventails, and demonstrate its application to event-based linear motion estimation, with known rotation from an Inertial Measurement Unit. Using this parametrization, we introduce a novel minimal 5-point solver that jointly estimates line parameters and linear camera velocity projections, which can be fused into a single, averaged linear velocity when considering multiple lines. We demonstrate on both synthetic and real data that our solver generates more stable relative motion estimates than other methods while capturing more inliers than clustering based on spatio-temporal planes. In particular, our method consistently achieves a 100% success rate in estimating linear velocity where existing closed-form solvers only achieve between 23% and 70%. The proposed eventails contribute to a better understanding of spatio-temporal event-generated geometries and we thus believe it will become a core building block of future event-based motion estimation algorithms.

Roberto Pellerito, Marco Cannici, Daniel Gehrig et al.

Visual Odometry (VO) is crucial for autonomous robotic navigation, especially in GPS-denied environments like planetary terrains. To improve robustness, recent model-based VO systems have begun combining standard and event-based cameras. While event cameras excel in low-light and high-speed motion, standard cameras provide dense and easier-to-track features. However, the field of image- and event-based VO still predominantly relies on model-based methods and is yet to fully integrate recent image-only advancements leveraging end-to-end learning-based architectures. Seamlessly integrating the two modalities remains challenging due to their different nature, one asynchronous, the other not, limiting the potential for a more effective image- and event-based VO. We introduce RAMP-VO, the first end-to-end learned image- and event-based VO system. It leverages novel Recurrent, Asynchronous, and Massively Parallel (RAMP) encoders capable of fusing asynchronous events with image data, providing 8x faster inference and 33% more accurate predictions than existing solutions. Despite being trained only in simulation, RAMP-VO outperforms previous methods on the newly introduced Apollo and Malapert datasets, and on existing benchmarks, where it improves image- and event-based methods by 58.8% and 30.6%, paving the way for robust and asynchronous VO in space.

Mariam Hassan, Sebastian Stapf, Ahmad Rahimi et al.

We present GEM, a Generalizable Ego-vision Multimodal world model that predicts future frames using a reference frame, sparse features, human poses, and ego-trajectories. Hence, our model has precise control over object dynamics, ego-agent motion and human poses. GEM generates paired RGB and depth outputs for richer spatial understanding. We introduce autoregressive noise schedules to enable stable long-horizon generations. Our dataset is comprised of 4000+ hours of multimodal data across domains like autonomous driving, egocentric human activities, and drone flights. Pseudo-labels are used to get depth maps, ego-trajectories, and human poses. We use a comprehensive evaluation framework, including a new Control of Object Manipulation (COM) metric, to assess controllability. Experiments show GEM excels at generating diverse, controllable scenarios and temporal consistency over long generations. Code, models, and datasets are fully open-sourced.

Roberto Pellerito, Nico Messikommer, Giovanni Cioffi et al.

In this work, we introduce the first framework for Motion-aware Event Suppression, which learns to filter events triggered by IMOs and ego-motion in real time. Our model jointly segments IMOs in the current event stream while predicting their future motion, enabling anticipatory suppression of dynamic events before they occur. Our lightweight architecture achieves 173 Hz inference on consumer-grade GPUs with less than 1 GB of memory usage, outperforming previous state-of-the-art methods on the challenging EVIMO benchmark by 67\% in segmentation accuracy while operating at a 53\% higher inference rate. Moreover, we demonstrate significant benefits for downstream applications: our method accelerates Vision Transformer inference by 83\% via token pruning and improves event-based visual odometry accuracy, reducing Absolute Trajectory Error (ATE) by 13\%.

Rong Zou, Marco Cannici, Davide Scaramuzza

Fast-flying aerial robots promise rapid inspection under limited battery constraints, with direct applications in infrastructure inspection, terrain exploration, and search and rescue. However, high speeds lead to severe motion blur in images and induce significant drift and noise in pose estimates, making dense 3D reconstruction with Neural Radiance Fields (NeRFs) particularly challenging due to their high sensitivity to such degradations. In this work, we present a unified framework that leverages asynchronous event streams alongside motion-blurred frames to reconstruct high-fidelity radiance fields from agile drone flights. By embedding event-image fusion into NeRF optimization and jointly refining event-based visual-inertial odometry priors using both event and frame modalities, our method recovers sharp radiance fields and accurate camera trajectories without ground-truth supervision. We validate our approach on both synthetic data and real-world sequences captured by a fast-flying drone. Despite highly dynamic drone flights, where RGB frames are severely degraded by motion blur and pose priors become unreliable, our method reconstructs high-fidelity radiance fields and preserves fine scene details, delivering a performance gain of over 50% on real-world data compared to state-of-the-art methods.

Vincenzo Polizzi, Marco Cannici, Davide Scaramuzza et al.

Camera relocalization methods range from dense image alignment to direct camera pose regression from a query image. Among these, sparse feature matching stands out as an efficient, versatile, and generally lightweight approach with numerous applications. However, feature-based methods often struggle with significant viewpoint and appearance changes, leading to matching failures and inaccurate pose estimates. To overcome this limitation, we propose a novel approach that leverages a globally sparse yet locally dense 3D representation of 2D features. By tracking and triangulating landmarks over a sequence of frames, we construct a sparse voxel map optimized to render image patch descriptors observed during tracking. Given an initial pose estimate, we first synthesize descriptors from the voxels using volumetric rendering and then perform feature matching to estimate the camera pose. This methodology enables the generation of descriptors for unseen views, enhancing robustness to view changes. We extensively evaluate our method on the 7-Scenes and Cambridge Landmarks datasets. Our results show that our method significantly outperforms existing state-of-the-art feature representation techniques in indoor environments, achieving up to a 39% improvement in median translation error. Additionally, our approach yields comparable results to other methods for outdoor scenarios while maintaining lower memory and computational costs.

Marco Cannici, Davide Scaramuzza

Neural Radiance Fields (NeRFs) have shown great potential in novel view synthesis. However, they struggle to render sharp images when the data used for training is affected by motion blur. On the other hand, event cameras excel in dynamic scenes as they measure brightness changes with microsecond resolution and are thus only marginally affected by blur. Recent methods attempt to enhance NeRF reconstructions under camera motion by fusing frames and events. However, they face challenges in recovering accurate color content or constrain the NeRF to a set of predefined camera poses, harming reconstruction quality in challenging conditions. This paper proposes a novel formulation addressing these issues by leveraging both model- and learning-based modules. We explicitly model the blur formation process, exploiting the event double integral as an additional model-based prior. Additionally, we model the event-pixel response using an end-to-end learnable response function, allowing our method to adapt to non-idealities in the real event-camera sensor. We show, on synthetic and real data, that the proposed approach outperforms existing deblur NeRFs that use only frames as well as those that combine frames and events by +6.13dB and +2.48dB, respectively.

Nico Messikommer, Jiaxu Xing, Leonard Bauersfeld et al.

Event cameras offer high temporal resolution and low latency, making them ideal sensors for high-speed robotic applications where conventional cameras suffer from image degradations such as motion blur. In addition, their low power consumption can enhance endurance, which is critical for resource-constrained platforms. Motivated by these properties, we present a novel approach that enables a quadrotor to fly through cluttered environments at high speed by perceiving the environment with a single event camera. Our proposed method employs an end-to-end neural network trained to map event data directly to control commands, eliminating the reliance on standard cameras. To enable efficient training in simulation, where rendering synthetic event data is computationally expensive, we propose Approximate Imitation Learning, a novel imitation learning framework. Our approach leverages a large-scale offline dataset to learn a task-specific representation space. Subsequently, the policy is trained through online interactions that rely solely on lightweight, simulated state information, eliminating the need to render events during training. This enables the efficient training of event-based control policies for fast quadrotor flight, highlighting the potential of our framework for other modalities where data simulation is costly or impractical. Our approach outperforms standard imitation learning baselines in simulation and demonstrates robust performance in real-world flight tests, achieving speeds up to 9.8 ms-1 in cluttered environments.

Manasi Muglikar, Nico Messikommer, Marco Cannici et al.

Adverse weather conditions, particularly heavy snowfall, pose significant challenges to both human drivers and autonomous vehicles. Traditional image-based de-snowing methods often introduce hallucination artifacts as they rely solely on spatial information, while video-based approaches require high frame rates and suffer from alignment artifacts at lower frame rates. Camera parameters, such as exposure time, also influence the appearance of snowflakes, making the problem difficult to solve and heavily dependent on network generalization. In this paper, we propose to address the challenge of desnowing by using event cameras, which offer compressed visual information with submillisecond latency, making them ideal for de-snowing images, even in the presence of ego-motion. Our method leverages the fact that snowflake occlusions appear with a very distinctive streak signature in the spatio-temporal representation of event data. We design an attention-based module that focuses on events along these streaks to determine when a background point was occluded and use this information to recover its original intensity. We benchmark our method on DSEC-Snow, a new dataset created using a green-screen technique that overlays pre-recorded snowfall data onto the existing DSEC driving dataset, resulting in precise ground truth and synchronized image and event streams. Our approach outperforms state-of-the-art de-snowing methods by 3 dB in PSNR for image reconstruction. Moreover, we show that off-the-shelf computer vision algorithms can be applied to our reconstructions for tasks such as depth estimation and optical flow, achieving a $20\%$ performance improvement over other de-snowing methods. Our work represents a crucial step towards enhancing the reliability and safety of vision systems in challenging winter conditions, paving the way for more robust, all-weather-capable applications.

Gokul Raju Govinda Raju, Nikola Zubić, Marco Cannici et al. · eth-zurich

With their motion-responsive nature, event-based cameras offer significant advantages over traditional cameras for optical flow estimation. While deep learning has improved upon traditional methods, current neural networks adopted for event-based optical flow still face temporal and spatial reasoning limitations. We propose Perturbed State Space Feature Encoders (P-SSE) for multi-frame optical flow with event cameras to address these challenges. P-SSE adaptively processes spatiotemporal features with a large receptive field akin to Transformer-based methods, while maintaining the linear computational complexity characteristic of SSMs. However, the key innovation that enables the state-of-the-art performance of our model lies in our perturbation technique applied to the state dynamics matrix governing the SSM system. This approach significantly improves the stability and performance of our model. We integrate P-SSE into a framework that leverages bi-directional flows and recurrent connections, expanding the temporal context of flow prediction. Evaluations on DSEC-Flow and MVSEC datasets showcase P-SSE's superiority, with 8.48% and 11.86% improvements in EPE performance, respectively.

Ivan Alberico, Marco Cannici, Giovanni Cioffi et al.

In this paper, we present a real-time egocentric trajectory prediction system for table tennis using event cameras. Unlike standard cameras, which suffer from high latency and motion blur at fast ball speeds, event cameras provide higher temporal resolution, allowing more frequent state updates, greater robustness to outliers, and accurate trajectory predictions using just a short time window after the opponent's impact. We collect a dataset of ping-pong game sequences, including 3D ground-truth trajectories of the ball, synchronized with sensor data from the Meta Project Aria glasses and event streams. Our system leverages foveated vision, using eye-gaze data from the glasses to process only events in the viewer's fovea. This biologically inspired approach improves ball detection performance and significantly reduces computational latency, as it efficiently allocates resources to the most perceptually relevant regions, achieving a reduction factor of 10.81 on the collected trajectories. Our detection pipeline has a worst-case total latency of 4.5 ms, including computation and perception - significantly lower than a frame-based 30 FPS system, which, in the worst case, takes 66 ms solely for perception. Finally, we fit a trajectory prediction model to the estimated states of the ball, enabling 3D trajectory forecasting in the future. To the best of our knowledge, this is the first approach to predict table tennis trajectories from an egocentric perspective using event cameras.

Pietro Bonazzi, Marie-Julie Rakatosaona, Marco Cannici et al.

Existing deep learning methods for the reconstruction and denoising of point clouds rely on small datasets of 3D shapes. We circumvent the problem by leveraging deep learning methods trained on billions of images. We propose a method to reconstruct point clouds from few images and to denoise point clouds from their rendering by exploiting prior knowledge distilled from image-based deep learning models. To improve reconstruction in constraint settings, we regularize the training of a differentiable renderer with hybrid surface and appearance by introducing semantic consistency supervision. In addition, we propose a pipeline to finetune Stable Diffusion to denoise renderings of noisy point clouds and we demonstrate how these learned filters can be used to remove point cloud noise coming without 3D supervision. We compare our method with DSS and PointRadiance and achieved higher quality 3D reconstruction on the Sketchfab Testset and SCUT Dataset.

Chiara Plizzari, Mirco Planamente, Gabriele Goletto et al.

Event cameras are novel bio-inspired sensors, which asynchronously capture pixel-level intensity changes in the form of "events". Due to their sensing mechanism, event cameras have little to no motion blur, a very high temporal resolution and require significantly less power and memory than traditional frame-based cameras. These characteristics make them a perfect fit to several real-world applications such as egocentric action recognition on wearable devices, where fast camera motion and limited power challenge traditional vision sensors. However, the ever-growing field of event-based vision has, to date, overlooked the potential of event cameras in such applications. In this paper, we show that event data is a very valuable modality for egocentric action recognition. To do so, we introduce N-EPIC-Kitchens, the first event-based camera extension of the large-scale EPIC-Kitchens dataset. In this context, we propose two strategies: (i) directly processing event-camera data with traditional video-processing architectures (E$^2$(GO)) and (ii) using event-data to distill optical flow information (E$^2$(GO)MO). On our proposed benchmark, we show that event data provides a comparable performance to RGB and optical flow, yet without any additional flow computation at deploy time, and an improved performance of up to 4% with respect to RGB only information.

Alberto Archetti, Marco Cannici, Matteo Matteucci

Recently, the trend of incorporating differentiable algorithms into deep learning architectures arose in machine learning research, as the fusion of neural layers and algorithmic layers has been beneficial for handling combinatorial data, such as shortest paths on graphs. Recent works related to data-driven planning aim at learning either cost functions or heuristic functions, but not both. We propose Neural Weighted A*, a differentiable anytime planner able to produce improved representations of planar maps as graph costs and heuristics. Training occurs end-to-end on raw images with direct supervision on planning examples, thanks to a differentiable A* solver integrated into the architecture. More importantly, the user can trade off planning accuracy for efficiency at run-time, using a single, real-valued parameter. The solution suboptimality is constrained within a linear bound equal to the optimal path cost multiplied by the tradeoff parameter. We experimentally show the validity of our claims by testing Neural Weighted A* against several baselines, introducing a novel, tile-based navigation dataset. We outperform similar architectures in planning accuracy and efficiency.

Mirco Planamente, Chiara Plizzari, Marco Cannici et al.

Event cameras are novel bio-inspired sensors, which asynchronously capture pixel-level intensity changes in the form of "events". The innovative way they acquire data presents several advantages over standard devices, especially in poor lighting and high-speed motion conditions. However, the novelty of these sensors results in the lack of a large amount of training data capable of fully unlocking their potential. The most common approach implemented by researchers to address this issue is to leverage simulated event data. Yet, this approach comes with an open research question: how well simulated data generalize to real data? To answer this, we propose to exploit, in the event-based context, recent Domain Adaptation (DA) advances in traditional computer vision, showing that DA techniques applied to event data help reduce the sim-to-real gap. To this purpose, we propose a novel architecture, which we call Multi-View DA4E (MV-DA4E), that better exploits the peculiarities of frame-based event representations while also promoting domain invariant characteristics in features. Through extensive experiments, we prove the effectiveness of DA methods and MV-DA4E on N-Caltech101. Moreover, we validate their soundness in a real-world scenario through a cross-domain analysis on the popular RGB-D Object Dataset (ROD), which we extended to the event modality (RGB-E).

Chiara Plizzari, Marco Cannici, Matteo Matteucci

Skeleton-based human action recognition has achieved a great interest in recent years, as skeleton data has been demonstrated to be robust to illumination changes, body scales, dynamic camera views, and complex background. Nevertheless, an effective encoding of the latent information underlying the 3D skeleton is still an open problem. In this work, we propose a novel Spatial-Temporal Transformer network (ST-TR) which models dependencies between joints using the Transformer self-attention operator. In our ST-TR model, a Spatial Self-Attention module (SSA) is used to understand intra-frame interactions between different body parts, and a Temporal Self-Attention module (TSA) to model inter-frame correlations. The two are combined in a two-stream network which outperforms state-of-the-art models using the same input data on both NTU-RGB+D 60 and NTU-RGB+D 120.

Chiara Plizzari, Marco Cannici, Matteo Matteucci

Skeleton-based Human Activity Recognition has achieved great interest in recent years as skeleton data has demonstrated being robust to illumination changes, body scales, dynamic camera views, and complex background. In particular, Spatial-Temporal Graph Convolutional Networks (ST-GCN) demonstrated to be effective in learning both spatial and temporal dependencies on non-Euclidean data such as skeleton graphs. Nevertheless, an effective encoding of the latent information underlying the 3D skeleton is still an open problem, especially when it comes to extracting effective information from joint motion patterns and their correlations. In this work, we propose a novel Spatial-Temporal Transformer network (ST-TR) which models dependencies between joints using the Transformer self-attention operator. In our ST-TR model, a Spatial Self-Attention module (SSA) is used to understand intra-frame interactions between different body parts, and a Temporal Self-Attention module (TSA) to model inter-frame correlations. The two are combined in a two-stream network, whose performance is evaluated on three large-scale datasets, NTU-RGB+D 60, NTU-RGB+D 120, and Kinetics Skeleton 400, consistently improving backbone results. Compared with methods that use the same input data, the proposed ST-TR achieves state-of-the-art performance on all datasets when using joints' coordinates as input, and results on-par with state-of-the-art when adding bones information.

Marco Cannici, Marco Ciccone, Andrea Romanoni et al.

Dynamic Vision Sensors (DVSs) asynchronously stream events in correspondence of pixels subject to brightness changes. Differently from classic vision devices, they produce a sparse representation of the scene. Therefore, to apply standard computer vision algorithms, events need to be integrated into a frame or event-surface. This is usually attained through hand-crafted grids that reconstruct the frame using ad-hoc heuristics. In this paper, we propose Matrix-LSTM, a grid of Long Short-Term Memory (LSTM) cells that efficiently process events and learn end-to-end task-dependent event-surfaces. Compared to existing reconstruction approaches, our learned event-surface shows good flexibility and expressiveness on optical flow estimation on the MVSEC benchmark and it improves the state-of-the-art of event-based object classification on the N-Cars dataset.

Marco Cannici, Marco Ciccone, Andrea Romanoni et al.

Event-based cameras are neuromorphic sensors capable of efficiently encoding visual information in the form of sparse sequences of events. Being biologically inspired, they are commonly used to exploit some of the computational and power consumption benefits of biological vision. In this paper we focus on a specific feature of vision: visual attention. We propose two attentive models for event based vision: an algorithm that tracks events activity within the field of view to locate regions of interest and a fully-differentiable attention procedure based on DRAW neural model. We highlight the strengths and weaknesses of the proposed methods on four datasets, the Shifted N-MNIST, Shifted MNIST-DVS, CIFAR10-DVS and N-Caltech101 collections, using the Phased LSTM recognition network as a baseline reference model obtaining improvements in terms of both translation and scale invariance.

Marco Cannici, Marco Ciccone, Andrea Romanoni et al.

Event-based cameras, also known as neuromorphic cameras, are bioinspired sensors able to perceive changes in the scene at high frequency with low power consumption. Becoming available only very recently, a limited amount of work addresses object detection on these devices. In this paper we propose two neural networks architectures for object detection: YOLE, which integrates the events into surfaces and uses a frame-based model to process them, and fcYOLE, an asynchronous event-based fully convolutional network which uses a novel and general formalization of the convolutional and max pooling layers to exploit the sparsity of camera events. We evaluate the algorithm with different extensions of publicly available datasets and on a novel synthetic dataset.