publications

2027

1. WACV

   

   Painting Change Map as Guidance for Generating Painting Processes

   M. Pobitzer, C. Liu, N. Biondi, and 1 more author

   In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, 2027

   arXiv Bib Code

   • Introduces the Painting Change Map, a spatio-temporal control signal that encodes where and when an artist acts across a 1,000-step continuum.
   • Enables fine-grained guidance of video generation for both forward painting and reverse “unpainting”.
   • Outperforms existing control baselines and is competitive with or better than state-of-the-art painting-process methods.

   @inproceedings{pobitzer2027painting,
     title = {Painting Change Map as Guidance for Generating Painting Processes},
     author = {Pobitzer, M. and Liu, C. and Biondi, N. and Sebe, N.},
     booktitle = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision},
     year = {2027},
   }

2026

1. TPAMI

   

   A Stationary (and Therefore Compatible) Representation is All You Need

   N. Biondi, F. Pernici, S. Ricci, and 1 more author

   IEEE Transactions on Pattern Analysis and Machine Intelligence, 2026

   Early access

   DOI arXiv Bib HTML Code Project Page

   • Proves that stationary representations learned with d-Simplex fixed classifiers imply compatibility under its formal definition.
   • Shows that combining cross-entropy with a contrastive loss captures higher-order dependencies and is equivalent to learning under the compatibility constraints.
   • Enables uninterrupted retrieval (no gallery re-indexing) while pre-trained models are fine-tuned and replaced, reaching state of the art.

   @article{biondi2024stationary_journal,
     title = {A Stationary (and Therefore Compatible) Representation is All You Need},
     author = {Biondi, N. and Pernici, F. and Ricci, S. and Del Bimbo, A.},
     journal = {IEEE Transactions on Pattern Analysis and Machine Intelligence},
     year = {2026},
     note = {Early access},
     doi = {10.1109/TPAMI.2026.3692279},
   }

2. NeurIPS

   

   DISTMOE: Rehearsal-free Routing in Mixture-of-Experts for Distributed Instruction Tuning

   M. Singha, N. Biondi, E. Ricci, and 1 more author

   In Advances in Neural Information Processing Systems, 2026

   arXiv Bib Code

   • Proposes DistMoE, a mixture-of-experts approach for distributed visual instruction tuning of MLLMs on private, domain-specific client data.
   • A public-anchored composition stage (routers + lightweight adapters, isotropic regularization) merges client experts without cross-client rehearsal.
   • At inference, token-wise routing over public and private experts composes domains without explicit domain labels.

   @inproceedings{singha2026distmoe,
     title = {{DISTMOE}: Rehearsal-free Routing in Mixture-of-Experts for Distributed Instruction Tuning},
     author = {Singha, M. and Biondi, N. and Ricci, E. and Roy, S.},
     booktitle = {Advances in Neural Information Processing Systems},
     year = {2026},
   }

3. NeurIPS

   

   Spherical Interpolation for Backward-Compatible Multimodal Representations

   S. Ricci, N. Biondi, and F. Pernici

   In Advances in Neural Information Processing Systems, 2026

   arXiv Bib Code

   • Tackles backward compatibility for vision-language models (CLIP/SigLIP) so a model upgrade does not require re-indexing the gallery.
   • Shows that spherical interpolation (SLERP) between the old query and the orthogonally aligned new query can land closer to the retrieval-optimal direction, with a margin-based Recall@K certification.
   • Recovers backward compatibility in most evaluated settings, improving over orthogonal alignment alone.

   @inproceedings{ricci2026spherical,
     title = {Spherical Interpolation for Backward-Compatible Multimodal Representations},
     author = {Ricci, S. and Biondi, N. and Pernici, F.},
     booktitle = {Advances in Neural Information Processing Systems},
     year = {2026},
   }

4. NeurIPS

   

   WASP: Weakly Aligned Spatiotemporal Pairs for Fetal Brain MRI-Ultrasound Learning

   F. Correnti, G. Magrini, M. Mistretta, and 6 more authors

   In Advances in Neural Information Processing Systems, 2026

   arXiv Bib Code

   • Teaches an ultrasound-only foundation model to understand fetal brain MRI, without subject-paired MRI–US scans.
   • Builds weak cross-modal pairs via entropic Optimal Transport on clinical metadata (gestational age and diagnostic planes) and fits a lightweight alignment module, with the backbone frozen.
   • On USFM, cuts gestational-age error from 21.9 → 17.4 days and raises plane classification from 61.9% → 69.0%.

   @inproceedings{correnti2026wasp,
     title = {{WASP}: Weakly Aligned Spatiotemporal Pairs for Fetal Brain {MRI}-Ultrasound Learning},
     author = {Correnti, F. and Magrini, G. and Mistretta, M. and Biondi, N. and Pala, P. and Ramalli, A. and Fiori, S. and Bagdanov, A. D. and Lenge, M.},
     booktitle = {Advances in Neural Information Processing Systems},
     year = {2026},
   }

5. ECCV WS

   Video Representation and Temporal Modeling for Shoplifting Detection in Retail Environments

   S. Ricci, N. Biondi, and A. Del Bimbo

   In European Conference on Computer Vision Workshops (RetailVision7), 2026

   Bib

   @inproceedings{ricci2026shoplifting,
     title = {Video Representation and Temporal Modeling for Shoplifting Detection in Retail Environments},
     author = {Ricci, S. and Biondi, N. and Del Bimbo, A.},
     booktitle = {European Conference on Computer Vision Workshops (RetailVision7)},
     year = {2026}
   }

6. IH&MMSec

   Improving Generalization in AI-Generated Facial Image Detection via Explainable Recovery

   G. Ciacci, A. Spinaci, N. Biondi, and 2 more authors

   In Proceedings of the ACM Workshop on Information Hiding and Multimedia Security, 2026

   Bib

   @inproceedings{ciacci2026improving,
     title = {Improving Generalization in {AI}-Generated Facial Image Detection via Explainable Recovery},
     author = {Ciacci, G. and Spinaci, A. and Biondi, N. and Ciamarra, A. and Caldelli, R.},
     booktitle = {Proceedings of the ACM Workshop on Information Hiding and Multimedia Security},
     year = {2026}
   }

7. CVPR

   

   PEPR: Privileged Event-based Predictive Regularization for Domain Generalization

   G. Magrini, F. Becattini, N. Biondi, and 1 more author

   In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2026

   Findings

   arXiv Bib Project Page

   • Uses event cameras as privileged information, available only at training time, to make a standalone RGB model robust to domain shift.
   • {“Reframes learning using privileged information as latent prediction”=>”the RGB encoder predicts event features instead of being forced to align with them, so it keeps its semantic detail.”}
   • Improves day-to-night and other domain shifts on both object detection and semantic segmentation.

   @inproceedings{magrini2026pepr,
     title = {{PEPR}: Privileged Event-based Predictive Regularization for Domain Generalization},
     author = {Magrini, G. and Becattini, F. and Biondi, N. and Pala, P.},
     booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
     note = {Findings},
     year = {2026},
   }

8. AAAI

   

   Mitigating Negative Flips via Margin Preserving Training

   S. Ricci, N. Biondi, F. Pernici, and 1 more author

   In Proceedings of the AAAI Conference on Artificial Intelligence, 2026

   arXiv Bib

   • {“Targets negative flips”=>”samples that the old model classified correctly and the updated model gets wrong after new classes are added.”}
   • Preserves the old model’s margins with an explicit logit margin-calibration term.
   • Adds a double-source focal distillation (old model + independently trained new model) so new classes stay accurate, reducing negative flips while keeping high overall accuracy.

   @inproceedings{ricci2026negative,
     title = {Mitigating Negative Flips via Margin Preserving Training},
     author = {Ricci, S. and Biondi, N. and Pernici, F. and Del Bimbo, A.},
     booktitle = {Proceedings of the AAAI Conference on Artificial Intelligence},
     volume = {40},
     number = {11},
     pages = {8721--8730},
     year = {2026},
   }

9. Smart Agri

   

   Parameter-efficient Vision Transformer adaptation for stem quality classification from smartphone forest images

   N. Biondi, S. Ricci, N. Arati, and 4 more authors

   Smart Agricultural Technology, 2026

   DOI Bib HTML

   • Classifies the stem quality of standing trees (Douglas fir) from ordinary smartphone forest images, enabling quality-aware digital forest inventories.
   • Benchmarks ViT backbones with k-NN, linear probing, full fine-tuning and LoRA; LoRA works best despite limited, imbalanced data.
   • Reaches ~0.78 accuracy at the forest-compartment level by majority voting over image-level predictions.

   @article{biondi2026parameter,
     title = {Parameter-efficient Vision Transformer adaptation for stem quality classification from smartphone forest images},
     author = {Biondi, N. and Ricci, S. and Arati, N. and Nocetti, M. and Aminti, G. and Pala, P. and Brunetti, M.},
     journal = {Smart Agricultural Technology},
     year = {2026},
     volume = {14},
     pages = {102132},
     doi = {10.1016/j.atech.2026.102132},
   }

2025

1. NeurIPS

   

   λ-Orthogonality Regularization for Compatible Representation Learning

   S. Ricci, N. Biondi, F. Pernici, and 2 more authors

   In Advances in Neural Information Processing Systems, 2025

   arXiv Bib Code

   • Introduces λ-Orthogonality, a relaxed orthogonality constraint on an affine adapter that maps new-model features into the old-model space.
   • Gets the adaptability of affine maps while preserving the geometry (and zero-shot ability) of the newly learned representation.
   • Validated across architectures and datasets for compatible retrieval-system updates.

   @inproceedings{ricci2025lambda,
     title = {$\lambda$-Orthogonality Regularization for Compatible Representation Learning},
     author = {Ricci, S. and Biondi, N. and Pernici, F. and Patras, I. and Del Bimbo, A.},
     booktitle = {Advances in Neural Information Processing Systems},
     year = {2025},
   }

2. ACM MM

   

   FRED: The Florence RGB-Event Drone Dataset

   G. Magrini, N. Marini, F. Becattini, and 4 more authors

   In Proceedings of the ACM International Conference on Multimedia, 2025

   arXiv Bib

   • Releases FRED, a synchronized RGB + event camera dataset for drone detection, tracking and trajectory forecasting.
   • Includes 7+ hours of densely annotated trajectories from 5 drone models, including rain and adverse lighting.
   • Provides standard evaluation protocols and metrics for reproducible benchmarking.

   @inproceedings{magrini2025fred,
     title = {{FRED}: The Florence {RGB}-Event Drone Dataset},
     author = {Magrini, G. and Marini, N. and Becattini, F. and Berlincioni, L. and Biondi, N. and Pala, P. and Del Bimbo, A.},
     booktitle = {Proceedings of the ACM International Conference on Multimedia},
     year = {2025},
   }

2024

1. CVPR

   

   Stationary Representations: Optimally Approximating Compatibility and Implications for Improved Model Replacements

   N. Biondi, F. Pernici, S. Ricci, and 1 more author

   In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024

   Highlight arXiv Bib HTML Code

   • Shows that stationary representations from a d-Simplex fixed classifier optimally approximate compatibility under both inequality constraints of its formal definition.
   • Studies the practical case of a model that is sequentially fine-tuned and asynchronously replaced by a better pre-trained model downloaded from elsewhere.
   • Delivers seamless retrieval service with improved accuracy and no gallery reprocessing. CVPR Highlight (top 2.8%).

   Highlight paper, top 2.8%

   @inproceedings{biondi2024stationary_cvpr,
     title = {Stationary Representations: Optimally Approximating Compatibility and Implications for Improved Model Replacements},
     author = {Biondi, N. and Pernici, F. and Ricci, S. and Del Bimbo, A.},
     booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
     pages = {28793--28804},
     year = {2024},
   }

2. ECCV WS

   

   ComiCap: A VLMs Pipeline for Dense Captioning of Comic Panels

   E. Vivoli, N. Biondi, M. Bertini, and 1 more author

   In European Conference on Computer Vision Workshops (AI4VA), 2024

   arXiv Bib Code

   • Builds a training-free VLM pipeline for dense, grounded captioning of comic panels.
   • Introduces an attribute-retaining metric and a densely annotated test set to select the best open-source VLM.
   • Annotates 2M+ panels from 13,000 books, outperforming specifically trained models.

   @inproceedings{vivoli2024comicap,
     title = {{ComiCap}: A VLMs Pipeline for Dense Captioning of Comic Panels},
     author = {Vivoli, E. and Biondi, N. and Bertini, M. and Karatzas, D.},
     booktitle = {European Conference on Computer Vision Workshops (AI4VA)},
     pages = {56--67},
     year = {2024},
   }

3. ECCV WS

   

   Backward-Compatible Aligned Representations via an Orthogonal Transformation Layer

   S. Ricci, N. Biondi, F. Pernici, and 1 more author

   In European Conference on Computer Vision Workshops (Beyond Euclidean), 2024

   arXiv Bib

   • Expands the representation with extra dimensions and learns an orthogonal transformation to stay compatible with the old model.
   • The orthogonal map preserves the old feature geometry while new dimensions absorb new knowledge, so no re-indexing is needed.
   • State-of-the-art compatibility and accuracy on CIFAR-100 and ImageNet-1k.

   @inproceedings{ricci2024backward,
     title = {Backward-Compatible Aligned Representations via an Orthogonal Transformation Layer},
     author = {Ricci, S. and Biondi, N. and Pernici, F. and Del Bimbo, A.},
     booktitle = {European Conference on Computer Vision Workshops (Beyond Euclidean)},
     year = {2024},
   }

4. ICDAR

   

   Comics Datasets Framework: Mix of Comics Datasets for Detection Benchmarking

   E. Vivoli, I. Campaioli, M. Nardoni, and 3 more authors

   In International Conference on Document Analysis and Recognition, 2024

   arXiv Bib Code

   • Unifies existing comics datasets into one annotation format with consistent splits and metrics.
   • Introduces Comics100, 100 books from the Digital Comics Museum, to counter the over-representation of manga.
   • Benchmarks several detection architectures, releasing code and weights for reproducible comparisons.

   @inproceedings{vivoli2024comics,
     title = {Comics Datasets Framework: Mix of Comics Datasets for Detection Benchmarking},
     author = {Vivoli, E. and Campaioli, I. and Nardoni, M. and Biondi, N. and Bertini, M. and Karatzas, D.},
     booktitle = {International Conference on Document Analysis and Recognition},
     pages = {154--167},
     year = {2024},
   }

5. TMLR

   Hyperspherical Simplex Representations from Softmax Outputs and Logits are Inherently Backward-Compatible

   S. Ricci, N. Biondi, F. Pernici, and 1 more author

   Transactions on Machine Learning Research, 2024

   Under review

   Bib

   @article{ricci2024hyperspherical,
     title = {Hyperspherical Simplex Representations from Softmax Outputs and Logits are Inherently Backward-Compatible},
     author = {Ricci, S. and Biondi, N. and Pernici, F. and Del Bimbo, A.},
     journal = {Transactions on Machine Learning Research},
     year = {2024},
     note = {Under review}
   }

2023

1. TPAMI

   

   CoReS: Compatible Representations via Stationarity

   N. Biondi, F. Pernici, M. Bruni, and 1 more author

   IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023

   DOI arXiv Bib HTML Code

   • Proposes CoReS, which learns compatible representations by relying on stationarity induced by polytope-based fixed classifiers.
   • Needs no mapping between old and new features and no pairwise training with the previous model.
   • Largely outperforms prior work, especially across multiple consecutive model upgrades.

   @article{biondi2021cores,
     title = {{CoReS}: Compatible Representations via Stationarity},
     author = {Biondi, N. and Pernici, F. and Bruni, M. and Del Bimbo, A.},
     journal = {IEEE Transactions on Pattern Analysis and Machine Intelligence},
     volume = {45},
     number = {8},
     pages = {9567--9582},
     year = {2023},
     doi = {10.1109/TPAMI.2023.3259542},
   }

2022

1. ICIAP

   

   Contrastive Supervised Distillation for Continual Representation Learning

   T. Barletti, N. Biondi, F. Pernici, and 2 more authors

   In International Conference on Image Analysis and Processing, 2022

   Best Student Paper DOI arXiv Bib HTML Code

   • {“Proposes Contrastive Supervised Distillation (CSD)”=>”the student is contrastively distilled from the teacher using label information, reducing feature forgetting.”}
   • Shows that feature forgetting in visual retrieval is less catastrophic than in classification.
   • Best Student Paper Award at ICIAP 2021.

   Best Student Paper Award (Oral)

   @inproceedings{barletti2022contrastive,
     title = {Contrastive Supervised Distillation for Continual Representation Learning},
     author = {Barletti, T. and Biondi, N. and Pernici, F. and Bruni, M. and Del Bimbo, A.},
     booktitle = {International Conference on Image Analysis and Processing},
     volume = {13231},
     pages = {597--609},
     year = {2022},
     doi = {10.1007/978-3-031-06427-2_50},
   }

2. ACM ToMM

   

   CL^2R: Compatible Lifelong Learning Representations

   N. Biondi, F. Pernici, M. Bruni, and 2 more authors

   ACM Transactions on Multimedia Computing, Communications, and Applications, 2022

   DOI arXiv Bib HTML Code

   • {“Formalizes Compatible Lifelong Learning Representations (CL²R)”=>”updating the model must never make gallery features unusable for visual search.”}
   • Identifies stationarity as the key property and enforces both local and global stationarity during training.
   • Introduces new metrics for evaluating compatibility under catastrophic forgetting.

   @article{biondi2021cl2r,
     title = {{CL$^2$R}: Compatible Lifelong Learning Representations},
     author = {Biondi, N. and Pernici, F. and Bruni, M. and Mugnai, D. and Del Bimbo, A.},
     journal = {ACM Transactions on Multimedia Computing, Communications, and Applications},
     volume = {18},
     number = {2s},
     pages = {132:1--132:22},
     year = {2022},
     doi = {10.1145/3564786},
   }