Thierry Chateau

Mehdi Gharbage, Céline Teulière, Pierre Bouges et al.

Vision foundation models pretrained on web-scale data have recently shown strong transfer capabilities on many downstream tasks, but their effectiveness for industrial visual inspection remains unclear. Industrial data differ substantially from web-data and often require fine-grained dense prediction, raising the question of whether modern self-supervised pretraining can improve over the conventional transfer-learning paradigm based on supervised ImageNet initialization. In this work, we compare ConvNeXt backbones pretrained with supervised ImageNet classification or DINOv3 distillation, and relate them to the conventional ResNet-50 baseline. We evaluate semantic segmentation, instance segmentation, and object detection across four downstream datasets spanning RGB surface-defect inspection and X-ray defect detection. We further study both frozen and fully finetuned adaptation regimes. Our results show that DINOv3 offers no clear advantage in frozen transfer, but provides a stronger initialization after full finetuning on RGB tasks, yielding faster convergence and better final performance. Under X-ray modality shift, however, supervised ImageNet pretraining remains more effective in both frozen and finetuned settings. Overall, our findings suggest that modern vision foundation models are promising for supervised RGB industrial inspection, but their transferability is strongly conditioned by downstream adaptation and target modality.

Andrew Caunes, Thierry Chateau, Vincent Fremont

Semantic and panoptic occupancy prediction for road scene analysis provides a dense 3D representation of the ego vehicle's surroundings. Current camera-only approaches typically rely on costly dense 3D supervision or require training models on data from the target domain, limiting deployment in unseen environments. We propose FreeOcc, a training-free pipeline that leverages pretrained foundation models to recover both semantics and geometry from multi-view images. FreeOcc extracts per-view panoptic priors with a promptable foundation segmentation model and prompt-to-taxonomy rules, and reconstructs metric 3D points with a reconstruction foundation model. Depth- and confidence- aware filtering lifts reliable labels into 3D, which are fused over time and voxelized with a deterministic refinement stack. For panoptic occupancy, instances are recovered by fitting and merging robust current-view 3D box candidates, enabling instance-aware occupancy without any learned 3D model. On Occ3D-nuScenes, FreeOcc achieves 16.9 mIoU and 16.5 RayIoU train-free, on par with state-of-the-art weakly supervised methods. When employed as a pseudo-label generation pipeline for training downstream models, it achieves 21.1 RayIoU, surpassing the previous state-of-the-art weakly supervised baseline. Furthermore, FreeOcc sets new baselines for both train-free and weakly supervised panoptic occupancy prediction, achieving 3.1 RayPQ and 3.9 RayPQ, respectively. These results highlight foundation-model-driven perception as a practical route to training-free 3D scene understanding.

Andrew Caunes, Thierry Chateau, Vincent Frémont

Semantic segmentation of 3D LiDAR point clouds, essential for autonomous driving and infrastructure management, is best achieved by supervised learning, which demands extensive annotated datasets and faces the problem of domain shifts. We introduce a new 3D semantic segmentation pipeline that leverages aligned scenes and state-of-the-art 2D segmentation methods, avoiding the need for direct 3D annotation or reliance on additional modalities such as camera images at inference time. Our approach generates 2D views from LiDAR scans colored by sensor intensity and applies 2D semantic segmentation to these views using a camera-domain pretrained model. The segmented 2D outputs are then back-projected onto the 3D points, with a simple voting-based estimator that merges the labels associated to each 3D point. Our main contribution is a global pipeline for 3D semantic segmentation requiring no prior 3D annotation and not other modality for inference, which can be used for pseudo-label generation. We conduct a thorough ablation study and demonstrate the potential of the generated pseudo-labels for the Unsupervised Domain Adaptation task.

Andrew Caunes, Thierry Chateau, Vincent Fremont

3D semantic segmentation is essential for autonomous driving and road infrastructure analysis, but state-of-the-art 3D models suffer from severe domain shift when applied across datasets. We propose a multi-view projection framework for unsupervised domain adaptation (UDA). Our method aligns LiDAR scans into coherent 3D scenes and renders them from multiple virtual camera poses to generate large-scale synthetic 2D datasets (PC2D) in various modalities. An ensemble of 2D segmentation models is trained on these modalities, and during inference, hundreds of views per scene are processed, with logits back-projected to 3D using an occlusion-aware voting scheme to produce point-wise labels. These labels can be used directly or to fine-tune a 3D segmentation model in the target domain. We evaluate our approach in both Real-to-Real and Simulation-to-Real UDA, achieving state-of-the-art performance in the Real-to-Real setting. Furthermore, we show that our framework enables segmentation of rare classes, leveraging only 2D annotations for those classes while relying on 3D annotations for others in the source domain.

Konaté Mohamed Abbas, Anne-Françoise Yao, Thierry Chateau et al.

In this paper, we investigate continual learning performance metrics used in class incremental learning strategies for continual learning (CL) using some high performing methods. We investigate especially mean task accuracy. First, we show that it lacks of expressiveness through some simple experiments to capture performance. We show that monitoring average tasks performance is over optimistic and can lead to misleading conclusions for future real life industrial uses. Then, we propose first a simple metric, Minimal Incremental Class Accuracy (MICA) which gives a fair and more useful evaluation of different continual learning methods. Moreover, in order to provide a simple way to easily compare different methods performance in continual learning, we derive another single scalar metric that take into account the learning performance variation as well as our newly introduced metric.

Ruddy Théodose, Dieumet Denis, Thierry Chateau et al.

Current neural networks-based object detection approaches processing LiDAR point clouds are generally trained from one kind of LiDAR sensors. However, their performances decrease when they are tested with data coming from a different LiDAR sensor than the one used for training, i.e., with a different point cloud resolution. In this paper, R-AGNO-RPN, a region proposal network built on fusion of 3D point clouds and RGB images is proposed for 3D object detection regardless of point cloud resolution. As our approach is designed to be also applied on low point cloud resolutions, the proposed method focuses on object localization instead of estimating refined boxes on reduced data. The resilience to low-resolution point cloud is obtained through image features accurately mapped to Bird's Eye View and a specific data augmentation procedure that improves the contribution of the RGB images. To show the proposed network's ability to deal with different point clouds resolutions, experiments are conducted on both data coming from the KITTI 3D Object Detection and the nuScenes datasets. In addition, to assess its performances, our method is compared to PointPillars, a well-known 3D detection network. Experimental results show that even on point cloud data reduced by $80\%$ of its original points, our method is still able to deliver relevant proposals localization.

Anthony Berthelier, Yongzhe Yan, Thierry Chateau et al.

While deep neural networks (DNNs) have proven to be efficient for numerous tasks, they come at a high memory and computation cost, thus making them impractical on resource-limited devices. However, these networks are known to contain a large number of parameters. Recent research has shown that their structure can be more compact without compromising their performance. In this paper, we present a sparsity-inducing regularization term based on the ratio l1/l2 pseudo-norm defined on the filter coefficients. By defining this pseudo-norm appropriately for the different filter kernels, and removing irrelevant filters, the number of kernels in each layer can be drastically reduced leading to very compact Deep Convolutional Neural Networks (DCNN) structures. Unlike numerous existing methods, our approach does not require an iterative retraining process and, using this regularization term, directly produces a sparse model during the training process. Furthermore, our approach is also much easier and simpler to implement than existing methods. Experimental results on MNIST and CIFAR-10 show that our approach significantly reduces the number of filters of classical models such as LeNet and VGG while reaching the same or even better accuracy than the baseline models. Moreover, the trade-off between the sparsity and the accuracy is compared to other loss regularization terms based on the l1 or l2 norm as well as the SSL, NISP and GAL methods and shows that our approach is outperforming them.

Yongzhe Yan, Stefan Duffner, Priyanka Phutane et al.

We present a facial landmark position correlation analysis as well as its applications. Although numerous facial landmark detection methods have been presented in the literature, few of them explicitly take into account the inherent relationship among landmarks. To reveal and interpret this relationship, we propose to analyze landmark correlation by using Canonical Correlation Analysis~(CCA). We experimentally show that the dense facial landmark annotations in current benchmarks are strongly correlated. We propose two applications based on this analysis. First, by analyzing the landmark correlation, we gain some interesting insights into the predictions of different landmark detection models (including random forests model and CNN models). We also demonstrate how CNNs progressively learn to predict facial landmarks. Second, we propose a few-shot learning method that allows to considerably reduce the manual effort for dense landmark annotation.

Yongzhe Yan, Stefan Duffner, Priyanka Phutane et al.

The recent performance of facial landmark detection has been significantly improved by using deep Convolutional Neural Networks (CNNs), especially the Heatmap Regression Models (HRMs). Although their performance on common benchmark datasets has reached a high level, the robustness of these models still remains a challenging problem in the practical use under noisy conditions of realistic environments. Contrary to most existing work focusing on the design of new models, we argue that improving the robustness requires rethinking many other aspects, including the use of datasets, the format of landmark annotation, the evaluation metric as well as the training and detection algorithm itself. In this paper, we propose a novel method for robust facial landmark detection, using a loss function based on the 2D Wasserstein distance combined with a new landmark coordinate sampling relying on the barycenter of the individual probability distributions. Our method can be plugged-and-play on most state-of-the-art HRMs with neither additional complexity nor structural modifications of the models. Further, with the large performance increase, we found that current evaluation metrics can no longer fully reflect the robustness of these models. Therefore, we propose several improvements to the standard evaluation protocol. Extensive experimental results on both traditional evaluation metrics and our evaluation metrics demonstrate that our approach significantly improves the robustness of state-of-the-art facial landmark detection models.

Ala Mhalla, Thierry Chateau, Houda Maamatou et al.

Generally, the performance of a generic detector decreases significantly when it is tested on a specific scene due to the large variation between the source training dataset and the samples from the target scene. To solve this problem, we propose a new formalism of transfer learning based on the theory of a Sequential Monte Carlo (SMC) filter to automatically specialize a scene-specific Faster R-CNN detector. The suggested framework uses different strategies based on the SMC filter steps to approximate iteratively the target distribution as a set of samples in order to specialize the Faster R-CNN detector towards a target scene. Moreover, we put forward a likelihood function that combines spatio-temporal information extracted from the target video sequence and the confidence-score given by the output layer of the Faster R-CNN, to favor the selection of target samples associated with the right label. The effectiveness of the suggested framework is demonstrated through experiments on several public traffic datasets. Compared with the state-of-the-art specialization frameworks, the proposed framework presents encouraging results for both single and multi-traffic object detections.

Florian Chabot, Mohamed Chaouch, Jaonary Rabarisoa et al.

In this paper, we present a novel approach, called Deep MANTA (Deep Many-Tasks), for many-task vehicle analysis from a given image. A robust convolutional network is introduced for simultaneous vehicle detection, part localization, visibility characterization and 3D dimension estimation. Its architecture is based on a new coarse-to-fine object proposal that boosts the vehicle detection. Moreover, the Deep MANTA network is able to localize vehicle parts even if these parts are not visible. In the inference, the network's outputs are used by a real time robust pose estimation algorithm for fine orientation estimation and 3D vehicle localization. We show in experiments that our method outperforms monocular state-of-the-art approaches on vehicle detection, orientation and 3D location tasks on the very challenging KITTI benchmark.