Pietro Cerri
Pietro Cerri has a total of 8 co-authors. Here are the top co-authors by number of shared publications:
- Fabio Pizzati
INRIA
4 shared publications - Raoul de Charette
INRIA
4 shared publications - Anthony Dell'Eva
University of Bologna
1 shared publication - Francesca Ghidini
Ambarella (United States)
1 shared publication - Marco Orsingher
University of Parma
1 shared publication - Massimo Bertozzi
University of Parma
1 shared publication
Latest Publications
- Open Access
- 4 July 2023
- Published by Springer Nature in Research Square
Real-Time Semantic Segmentation of Spherical Images for Automotive Applications
Recent advancements in autonomous driving technology have resulted in a growing need for robust algorithms that can effectively detect, recognize, and segment objects in the surrounding environment. Semantic segmentation systems, which classify each pixel of an image into a class, have become a key component in achieving complete scene understanding, a crucial task for self-driving vehicles. This work focuses on utilizing deep learning methods to perform semantic segmentation on spherical images, a particular representation of fisheye images, characterized by a large field of view and strong distortion, which can result in performance degradation on convolutional neural networks (CNNs) trained on pinhole images. To address this issue, we propose the use of spherical kernels, which adapt their sampling locations according to the spherical projection, as an alternative to standard convolutional kernels, in real-time CNNs. We demonstrate that this approach outperforms the use of the regular convolution operation. To showcase the effectiveness of this method, we generate three spherical image datasets with varying characteristics.
- 4 July 2023
- Published by Springer Nature in Research Square
Open AccessReal-Time Semantic Segmentation of Spherical Images for Automotive Applications
Recent advancements in autonomous driving technology have resulted in a growing need for robust algorithms that can effectively detect, recognize, and segment objects in the surrounding environment. Semantic segmentation systems, which classify each pixel of an image into a class, have become a key component in achieving complete scene understanding, a crucial task for self-driving vehicles. This work focuses on utilizing deep learning methods to perform semantic segmentation on spherical images, a particular representation of fisheye images, characterized by a large field of view and strong distortion, which can result in performance degradation on convolutional neural networks (CNNs) trained on pinhole images. To address this issue, we propose the use of spherical kernels, which adapt their sampling locations according to the spherical projection, as an alternative to standard convolutional kernels, in real-time CNNs. We demonstrate that this approach outperforms the use of the regular convolution operation. To showcase the effectiveness of this method, we generate three spherical image datasets with varying characteristics.
- Open Access
Physics-informed Guided Disentanglement in Generative Networks
Image-to-image translation (i2i) networks suffer from entanglement effects in presence of physics-related phenomena in target domain (such as occlusions, fog, etc), lowering altogether the translation quality, controllability and variability. In this paper, we propose a general framework to disentangle visual traits in target images. Primarily, we build upon collection of simple physics models, guiding the disentanglement with a physical model that renders some of the target traits, and learning the remaining ones. Because physics allows explicit and interpretable outputs, our physical models (optimally regressed on target) allows generating unseen scenarios in a controllable manner. Secondarily, we show the versatility of our framework to neural-guided disentanglement where a generative network is used in place of a physical model in case the latter is not directly accessible. Altogether, we introduce three strategies of disentanglement being guided from either a fully differentiable physics model, a (partially) non-differentiable physics model, or a neural network. The results show our disentanglement strategies dramatically increase performances qualitatively and quantitatively in several challenging scenarios for image translation.
Open AccessPhysics-informed Guided Disentanglement in Generative Networks
Image-to-image translation (i2i) networks suffer from entanglement effects in presence of physics-related phenomena in target domain (such as occlusions, fog, etc), lowering altogether the translation quality, controllability and variability. In this paper, we propose a general framework to disentangle visual traits in target images. Primarily, we build upon collection of simple physics models, guiding the disentanglement with a physical model that renders some of the target traits, and learning the remaining ones. Because physics allows explicit and interpretable outputs, our physical models (optimally regressed on target) allows generating unseen scenarios in a controllable manner. Secondarily, we show the versatility of our framework to neural-guided disentanglement where a generative network is used in place of a physical model in case the latter is not directly accessible. Altogether, we introduce three strategies of disentanglement being guided from either a fully differentiable physics model, a (partially) non-differentiable physics model, or a neural network. The results show our disentanglement strategies dramatically increase performances qualitatively and quantitatively in several challenging scenarios for image translation.
- Open Access
CoMoGAN: continuous model-guided image-to-image translation
CoMoGAN is a continuous GAN relying on the unsupervised reorganization of the target data on a functional manifold. To that matter, we introduce a new Functional Instance Normalization layer and residual mechanism, which together disentangle image content from position on target manifold. We rely on naive physics-inspired models to guide the training while allowing private model/translations features. CoMoGAN can be used with any GAN backbone and allows new types of image translation, such as cyclic image translation like timelapse generation, or detached linear translation. On all datasets, it outperforms the literature. Our code is available at http://github.com/cv-rits/CoMoGAN .
Open AccessCoMoGAN: continuous model-guided image-to-image translation
CoMoGAN is a continuous GAN relying on the unsupervised reorganization of the target data on a functional manifold. To that matter, we introduce a new Functional Instance Normalization layer and residual mechanism, which together disentangle image content from position on target manifold. We rely on naive physics-inspired models to guide the training while allowing private model/translations features. CoMoGAN can be used with any GAN backbone and allows new types of image translation, such as cyclic image translation like timelapse generation, or detached linear translation. On all datasets, it outperforms the literature. Our code is available at http://github.com/cv-rits/CoMoGAN .
- Open Access
Model-based occlusion disentanglement for image-to-image translation
Image-to-image translation is affected by entanglement phenomena, which may occur in case of target data encompassing occlusions such as raindrops, dirt, etc. Our unsupervised model-based learning disentangles scene and occlusions, while benefiting from an adversarial pipeline to regress physical parameters of the occlusion model. The experiments demonstrate our method is able to handle varying types of occlusions and generate highly realistic translations, qualitatively and quantitatively outperforming the state-of-the-art on multiple datasets.
Open AccessModel-based occlusion disentanglement for image-to-image translation
Image-to-image translation is affected by entanglement phenomena, which may occur in case of target data encompassing occlusions such as raindrops, dirt, etc. Our unsupervised model-based learning disentangles scene and occlusions, while benefiting from an adversarial pipeline to regress physical parameters of the occlusion model. The experiments demonstrate our method is able to handle varying types of occlusions and generate highly realistic translations, qualitatively and quantitatively outperforming the state-of-the-art on multiple datasets.
- Open Access
Domain Bridge for Unpaired Image-to-Image Translation and Unsupervised Domain Adaptation
Image-to-image translation architectures may have limited effectiveness in some circumstances. For example, while generating rainy scenarios, they may fail to model typical traits of rain as water drops, and this ultimately impacts the synthetic images realism. With our method, called domain bridge, web-crawled data are exploited to reduce the domain gap, leading to the inclusion of previously ignored elements in the generated images. We make use of a network for clear to rain translation trained with the domain bridge to extend our work to Unsupervised Domain Adaptation (UDA). In that context, we introduce an online multimodal style-sampling strategy, where image translation multimodality is exploited at training time to improve performances. Finally, a novel approach for self-supervised learning is presented, and used to further align the domains. With our contributions, we simultaneously increase the realism of the generated images, while reaching on par performances with respect to the UDA state-of-the-art, with a simpler approach.
Open AccessDomain Bridge for Unpaired Image-to-Image Translation and Unsupervised Domain Adaptation
Image-to-image translation architectures may have limited effectiveness in some circumstances. For example, while generating rainy scenarios, they may fail to model typical traits of rain as water drops, and this ultimately impacts the synthetic images realism. With our method, called domain bridge, web-crawled data are exploited to reduce the domain gap, leading to the inclusion of previously ignored elements in the generated images. We make use of a network for clear to rain translation trained with the domain bridge to extend our work to Unsupervised Domain Adaptation (UDA). In that context, we introduce an online multimodal style-sampling strategy, where image translation multimodality is exploited at training time to improve performances. Finally, a novel approach for self-supervised learning is presented, and used to further align the domains. With our contributions, we simultaneously increase the realism of the generated images, while reaching on par performances with respect to the UDA state-of-the-art, with a simpler approach.