arXiv subjects

Alexandru Paul Condurache

Publications and source records attributed to Alexandru Paul Condurache.

At least 19 recordsLinked to original sources

arXiv2026

Towards Foundation Models for 3D Scene Understanding: Instance-Aware Self-Supervised Learning for Point Clouds

Recent advances in self-supervised learning (SSL) for point clouds have substantially improved 3D scene understanding without human annotations. Existing approaches emphasize semantic awareness by enforcing feature consistency across augmented views or by masked scene modeling. However, the resulting representations transfer poorly to instance localization, and often require full finetuning for strong performance. Instance awareness is a fundamental component of 3D perception, thus bridging this gap is crucial for progressing toward true 3D foundation models that support all downstream tasks on 3D data. In this work, we introduce PointINS, an instance-oriented self-supervised framework that enriches point cloud representations through geometry-aware learning. PointINS employs an orthogonal offset branch to jointly learn high-level semantic understanding and geometric reasoning, yielding instance awareness. We identify two consistent properties essential for robust instance localization and formulate them as complementary regularization strategies, Offset Distribution Regularization (ODR), which aligns predicted offsets with empirically observed geometric priors, and Spatial Clustering Regularization (SCR), which enforces local coherence by regularizing offsets with pseudo-instance masks. Through extensive experiments across five datasets, PointINS achieves on average +3.5% mAP improvement for indoor instance segmentation and +4.1% PQ gain for outdoor panoptic segmentation, paving the way for scalable 3D foundation models.

cs.CV↗

arXiv2026

Driving on Memory

End-to-end autonomous driving models plan future trajectories from raw sensor input. While earlier driving benchmarks often measured deviation from the human trajectory, current benchmarks such as NAVSIM and Bench2Drive evaluate models with richer simulation-based metrics intended to capture safe and compliant driving. A high benchmark score should reflect that a model can understand the scene in front of it and act accordingly. But how much of that score specifically comes from reacting to the dynamic part of that scene? To probe this, we remove a model's camera input and replace it with memories from prior drives at the same location. The retrieved memories can provide persistent scene information, including road layout and location-conditioned regularities, but not the current traffic state. Surprisingly, memory is nearly sufficient on NAVSIM, reaching or even exceeding the performance of leading end-to-end methods without actually observing the evaluated scene. Our results suggest that a high NAVSIM score does not require a planner to react to the current traffic scene and should be treated with caution. This effect is benchmark-dependent: driving from memory causes substantially larger performance drops on Bench2Drive and RealEngine. We provide our code at https://github.com/boschresearch/MemoryDrivoR .

cs.CV↗

arXiv2026

GhostPoint: Self-Supervised Representation Learning by Hallucinating Occluded LiDAR Structure

3D object detection from LiDAR point clouds is a core problem in autonomous driving. Recent advances in self-supervised learning (SSL) enable scalable pretraining and transfers well to per-point tasks such as semantic and panoptic segmentation, but transfer to 3D detection remains weaker. We analyze recent SSL methods and find that most objectives are defined only on measured LiDAR returns from visible surfaces, leaving occluded and unobserved regions unconstrained. This visible-surface bias can be sufficient for point-wise prediction, but 3D detection requires robustness to missing structure. To address this gap, we propose GhostPoint, an SSL framework that hallucinates latent features in local neighborhoods around discovered instances, generated via a novel instance voxel dilation. In GhostPoint, an encoder processes observed returns, and an additional predictor infers neighborhood representations from observed context. In addition to standard encoder-level supervision, we introduce a predictor-level supervision scheme on sampled voxels from generated neighborhoods. Specifically, observed (visible/masked) voxels match teacher-encoder targets, while unobserved voxels match teacher-predictor hallucinations. This design encourages the learned representation to explicitly model structure beyond observed returns. Extensive evaluations on nuScenes and Waymo demonstrate that our method achieves state-of-the-art performance, consistently improving downstream 3D detection, especially under sparse scans and limited labels.

cs.CV↗

arXiv2026

Unsupervised Point Cloud Registration with Self-Distillation

Rigid point cloud registration is a fundamental problem and highly relevant in robotics and autonomous driving. Nowadays deep learning methods can be trained to match a pair of point clouds, given the transformation between them. However, this training is often not scalable due to the high cost of collecting ground truth poses. Therefore, we present a self-distillation approach to learn point cloud registration in an unsupervised fashion. Here, each sample is passed to a teacher network and an augmented view is passed to a student network. The teacher includes a trainable feature extractor and a learning-free robust solver such as RANSAC. The solver forces consistency among correspondences and optimizes for the unsupervised inlier ratio, eliminating the need for ground truth labels. Our approach simplifies the training procedure by removing the need for initial hand-crafted features or consecutive point cloud frames as seen in related methods. We show that our method not only surpasses them on the RGB-D benchmark 3DMatch but also generalizes well to automotive radar, where classical features adopted by others fail. The code is available at https://github.com/boschresearch/direg .

cs.CV↗

arXiv2026

HASTE: A Framework for Training-Free, Dynamic, and Steerable Compression of Pre-Trained Convolutional Neural Networks

Deploying large convolutional neural networks (CNNs) on resource-constrained devices is challenging due to their high computational cost. While dynamic execution methods are promising, existing approaches for CNNs typically require specialized training or fine-tuning, limiting their effectiveness when applied to pre-trained models and requiring data access. To address this gap, we propose HASTE (Hashing for Tractable Efficiency), a plug-and-play convolution module that enables training-free, dynamic compression of large pre-trained CNNs. At inference time, HASTE uses locality-sensitive hashing to identify and merge redundant channels of latent feature maps on a patch-wise basis. This process simultaneously compresses the depth of both input features and their corresponding filters, resulting in computationally cheaper convolutions. We conduct extensive experiments on CIFAR-10 and ImageNet across a range of architectures, demonstrating a 46.2% FLOPs reduction in a ResNet34 on CIFAR-10 with only a 1.25% drop in accuracy, without any retraining. We support our claims by comprehensive ablation studies to validate our core design choices, an analysis of the method's properties and limitations, and a discussion that connects our channel merging scheme to the conceptually related task of token merging in Vision Transformers. Our results demonstrate that HASTE provides an effective solution for steerable compression of pre-trained CNNs at runtime, opening new possibilities for the deployment of efficient deep learning methods.

cs.CV↗

arXiv2026

Collaborative Learning for Semi-Supervised LiDAR Semantic Segmentation

Annotating large-scale LiDAR point clouds for 3D semantic segmentation is costly and time-consuming, which motivates the use of semi-supervised learning (SemiSL). Standard LiDAR SemiSL methods typically adopt a two-step training paradigm, where pseudo-labels are separately generated from a single distillation source, either from the same or another LiDAR representation. Such supervision relies on a unique source of pseudo-labels, which can reinforce confirmation bias and propagate errors during training, ultimately limiting performance. To address this challenge, we introduce CoLLiS, a novel framework that leverages Collaborative Learning for LiDAR Semi-supervised segmentation. Unlike prior paradigms with decoupled pseudo-labeling and training phases, CoLLiS trains multiple representations collaboratively in a single step by treating them as coequal students. Each student is adaptively distilled from multiple representations, while inter-student disparities are monitored online to resolve contradictory supervision and effectively mitigate confirmation bias. Extensive experiments on three datasets demonstrate that CoLLiS consistently outperforms state-of-the-art LiDAR SemiSL methods, with particularly strong gains in low-label regimes.

cs.CV↗

arXiv2026

Variance-Based Pruning for Accelerating and Compressing Trained Networks

Increasingly expensive training of ever larger models such as Vision Transfomers motivate reusing the vast library of already trained state-of-the-art networks. However, their latency, high computational costs and memory demands pose significant challenges for deployment, especially on resource-constrained hardware. While structured pruning methods can reduce these factors, they often require costly retraining, sometimes for up to hundreds of epochs, or even training from scratch to recover the lost accuracy resulting from the structural modifications. Maintaining the provided performance of trained models after structured pruning and thereby avoiding extensive retraining remains a challenge. To solve this, we introduce Variance-Based Pruning, a simple and structured one-shot pruning technique for efficiently compressing networks, with minimal finetuning. Our approach first gathers activation statistics, which are used to select neurons for pruning. Simultaneously the mean activations are integrated back into the model to preserve a high degree of performance. On ImageNet-1k recognition tasks, we demonstrate that directly after pruning DeiT-Base retains over 70% of its original performance and requires only 10 epochs of fine-tuning to regain 99% of the original accuracy while simultaneously reducing MACs by 35% and model size by 36%, thus speeding up the model by 1.44x. The code is available at: https://github.com/boschresearch/variance-based-pruning

cs.CV↗

arXiv2026

SparseLaneSTP: Leveraging Spatio-Temporal Priors with Sparse Transformers for 3D Lane Detection

3D lane detection has emerged as a critical challenge in autonomous driving, encompassing identification and localization of lane markings and the 3D road surface. Conventional 3D methods detect lanes from dense birds-eye-viewed (BEV) features, though erroneous transformations often result in a poor feature representation misaligned with the true 3D road surface. While recent sparse lane detectors have surpassed dense BEV approaches, they completely disregard valuable lane-specific priors. Furthermore, existing methods fail to utilize historic lane observations, which yield the potential to resolve ambiguities in situations of poor visibility. To address these challenges, we present SparseLaneSTP, a novel method that integrates both geometric properties of the lane structure and temporal information into a sparse lane transformer. It introduces a new lane-specific spatio-temporal attention mechanism, a continuous lane representation tailored for sparse architectures as well as temporal regularization. Identifying weaknesses of existing 3D lane datasets, we also introduce a precise and consistent 3D lane dataset using a simple yet effective auto-labeling strategy. Our experimental section proves the benefits of our contributions and demonstrates state-of-the-art performance across all detection and error metrics on existing 3D lane detection benchmarks as well as on our novel dataset.

cs.CV↗

arXiv2025

FLARES: Fast and Accurate LiDAR Multi-Range Semantic Segmentation

3D scene understanding is a critical yet challenging task in autonomous driving due to the irregularity and sparsity of LiDAR data, as well as the computational demands of processing large-scale point clouds. Recent methods leverage range-view representations to enhance efficiency, but they often adopt higher azimuth resolutions to mitigate information loss during spherical projection, where only the closest point is retained for each 2D grid. However, processing wide panoramic range-view images remains inefficient and may introduce additional distortions. Our empirical analysis shows that training with multiple range images, obtained from splitting the full point cloud, improves both segmentation accuracy and computational efficiency. However, this approach also poses new challenges of exacerbated class imbalance and increase in projection artifacts. To address these, we introduce FLARES, a novel training paradigm that incorporates two tailored data augmentation techniques and a specialized post-processing method designed for multi-range settings. Extensive experiments demonstrate that FLARES is highly generalizable across different architectures, yielding 2.1%~7.9% mIoU improvements on SemanticKITTI and 1.8%~3.9% mIoU on nuScenes, while delivering over 40% speed-up in inference.

cs.CV↗

arXiv2025

PseudoMapTrainer: Learning Online Mapping without HD Maps

Online mapping models show remarkable results in predicting vectorized maps from multi-view camera images only. However, all existing approaches still rely on ground-truth high-definition maps during training, which are expensive to obtain and often not geographically diverse enough for reliable generalization. In this work, we propose PseudoMapTrainer, a novel approach to online mapping that uses pseudo-labels generated from unlabeled sensor data. We derive those pseudo-labels by reconstructing the road surface from multi-camera imagery using Gaussian splatting and semantics of a pre-trained 2D segmentation network. In addition, we introduce a mask-aware assignment algorithm and loss function to handle partially masked pseudo-labels, allowing for the first time the training of online mapping models without any ground-truth maps. Furthermore, our pseudo-labels can be effectively used to pre-train an online model in a semi-supervised manner to leverage large-scale unlabeled crowdsourced data. The code is available at github.com/boschresearch/PseudoMapTrainer.

cs.CV↗

arXiv2025

PROM: Prioritize Reduction of Multiplications Over Lower Bit-Widths for Efficient CNNs

Convolutional neural networks (CNNs) are crucial for computer vision tasks on resource-constrained devices. Quantization effectively compresses these models, reducing storage size and energy cost. However, in modern depthwise-separable architectures, the computational cost is distributed unevenly across its components, with pointwise operations being the most expensive. By applying a general quantization scheme to this imbalanced cost distribution, existing quantization approaches fail to fully exploit potential efficiency gains. To this end, we introduce PROM, a straightforward approach for quantizing modern depthwise-separable convolutional networks by selectively using two distinct bit-widths. Specifically, pointwise convolutions are quantized to ternary weights, while the remaining modules use 8-bit weights, which is achieved through a simple quantization-aware training procedure. Additionally, by quantizing activations to 8-bit, our method transforms pointwise convolutions with ternary weights into int8 additions, which enjoy broad support across hardware platforms and effectively eliminates the need for expensive multiplications. Applying PROM to MobileNetV2 reduces the model's energy cost by more than an order of magnitude (23.9x) and its storage size by 2.7x compared to the float16 baseline while retaining similar classification performance on ImageNet. Our method advances the Pareto frontier for energy consumption vs. top-1 accuracy for quantized convolutional models on ImageNet. PROM addresses the challenges of quantizing depthwise-separable convolutional networks to both ternary and 8-bit weights, offering a simple way to reduce energy cost and storage size.

cs.CV↗

arXiv2025

Efficient Data Driven Mixture-of-Expert Extraction from Trained Networks

Vision Transformers have emerged as the state-of-the-art models in various Computer Vision tasks, but their high computational and resource demands pose significant challenges. While Mixture-of-Experts (MoE) can make these models more efficient, they often require costly retraining or even training from scratch. Recent developments aim to reduce these computational costs by leveraging pretrained networks. These have been shown to produce sparse activation patterns in the Multi-Layer Perceptrons (MLPs) of the encoder blocks, allowing for conditional activation of only relevant subnetworks for each sample. Building on this idea, we propose a new method to construct MoE variants from pretrained models. Our approach extracts expert subnetworks from the model's MLP layers post-training in two phases. First, we cluster output activations to identify distinct activation patterns. In the second phase, we use these clusters to extract the corresponding subnetworks responsible for producing them. On ImageNet-1k recognition tasks, we demonstrate that these extracted experts can perform surprisingly well out of the box and require only minimal fine-tuning to regain 98% of the original performance, all while reducing MACs and model size, by up to 36% and 32% respectively.

cs.CV↗

arXiv2025

Data-Free Dynamic Compression of CNNs for Tractable Efficiency

To reduce the computational cost of convolutional neural networks (CNNs) on resource-constrained devices, structured pruning approaches have shown promise in lowering floating-point operations (FLOPs) without substantial drops in accuracy. However, most methods require fine-tuning or specific training procedures to achieve a reasonable trade-off between retained accuracy and reduction in FLOPs, adding computational overhead and requiring training data to be available. To this end, we propose HASTE (Hashing for Tractable Efficiency), a data-free, plug-and-play convolution module that instantly reduces a network's test-time inference cost without training or fine-tuning. Our approach utilizes locality-sensitive hashing (LSH) to detect redundancies in the channel dimension of latent feature maps, compressing similar channels to reduce input and filter depth simultaneously, resulting in cheaper convolutions. We demonstrate our approach on the popular vision benchmarks CIFAR-10 and ImageNet, where we achieve a 46.72% reduction in FLOPs with only a 1.25% loss in accuracy by swapping the convolution modules in a ResNet34 on CIFAR-10 for our HASTE module.

cs.CV↗

arXiv2023

Dimensionality Reduced Training by Pruning and Freezing Parts of a Deep Neural Network, a Survey

State-of-the-art deep learning models have a parameter count that reaches into the billions. Training, storing and transferring such models is energy and time consuming, thus costly. A big part of these costs is caused by training the network. Model compression lowers storage and transfer costs, and can further make training more efficient by decreasing the number of computations in the forward and/or backward pass. Thus, compressing networks also at training time while maintaining a high performance is an important research topic. This work is a survey on methods which reduce the number of trained weights in deep learning models throughout the training. Most of the introduced methods set network parameters to zero which is called pruning. The presented pruning approaches are categorized into pruning at initialization, lottery tickets and dynamic sparse training. Moreover, we discuss methods that freeze parts of a network at its random initialization. By freezing weights, the number of trainable parameters is shrunken which reduces gradient computations and the dimensionality of the model's optimization space. In this survey we first propose dimensionality reduced training as an underlying mathematical model that covers pruning and freezing during training. Afterwards, we present and discuss different dimensionality reduced training methods.

cs.LG↗

arXiv2023

Boosting Deep Neural Networks with Geometrical Prior Knowledge: A Survey

Deep Neural Networks achieve state-of-the-art results in many different problem settings by exploiting vast amounts of training data. However, collecting, storing and - in the case of supervised learning - labelling the data is expensive and time-consuming. Additionally, assessing the networks' generalization abilities or predicting how the inferred output changes under input transformations is complicated since the networks are usually treated as a black box. Both of these problems can be mitigated by incorporating prior knowledge into the neural network. One promising approach, inspired by the success of convolutional neural networks in computer vision tasks, is to incorporate knowledge about symmetric geometrical transformations of the problem to solve that affect the output in a predictable way. This promises an increased data efficiency and more interpretable network outputs. In this survey, we try to give a concise overview about different approaches that incorporate geometrical prior knowledge into neural networks. Additionally, we connect those methods to 3D object detection for autonomous driving, where we expect promising results when applying those methods.

cs.CV↗

arXiv2023

Deep Neural Networks with Efficient Guaranteed Invariances

We address the problem of improving the performance and in particular the sample complexity of deep neural networks by enforcing and guaranteeing invariances to symmetry transformations rather than learning them from data. Group-equivariant convolutions are a popular approach to obtain equivariant representations. The desired corresponding invariance is then imposed using pooling operations. For rotations, it has been shown that using invariant integration instead of pooling further improves the sample complexity. In this contribution, we first expand invariant integration beyond rotations to flips and scale transformations. We then address the problem of incorporating multiple desired invariances into a single network. For this purpose, we propose a multi-stream architecture, where each stream is invariant to a different transformation such that the network can simultaneously benefit from multiple invariances. We demonstrate our approach with successful experiments on Scaled-MNIST, SVHN, CIFAR-10 and STL-10.

cs.LG↗

arXiv2022

Interspace Pruning: Using Adaptive Filter Representations to Improve Training of Sparse CNNs

Unstructured pruning is well suited to reduce the memory footprint of convolutional neural networks (CNNs), both at training and inference time. CNNs contain parameters arranged in $K \times K$ filters. Standard unstructured pruning (SP) reduces the memory footprint of CNNs by setting filter elements to zero, thereby specifying a fixed subspace that constrains the filter. Especially if pruning is applied before or during training, this induces a strong bias. To overcome this, we introduce interspace pruning (IP), a general tool to improve existing pruning methods. It uses filters represented in a dynamic interspace by linear combinations of an underlying adaptive filter basis (FB). For IP, FB coefficients are set to zero while un-pruned coefficients and FBs are trained jointly. In this work, we provide mathematical evidence for IP's superior performance and demonstrate that IP outperforms SP on all tested state-of-the-art unstructured pruning methods. Especially in challenging situations, like pruning for ImageNet or pruning to high sparsity, IP greatly exceeds SP with equal runtime and parameter costs. Finally, we show that advances of IP are due to improved trainability and superior generalization ability.

cs.CV↗

arXiv2022

Improving the Sample-Complexity of Deep Classification Networks with Invariant Integration

Leveraging prior knowledge on intraclass variance due to transformations is a powerful method to improve the sample complexity of deep neural networks. This makes them applicable to practically important use-cases where training data is scarce. Rather than being learned, this knowledge can be embedded by enforcing invariance to those transformations. Invariance can be imposed using group-equivariant convolutions followed by a pooling operation. For rotation-invariance, previous work investigated replacing the spatial pooling operation with invariant integration which explicitly constructs invariant representations. Invariant integration uses monomials which are selected using an iterative approach requiring expensive pre-training. We propose a novel monomial selection algorithm based on pruning methods to allow an application to more complex problems. Additionally, we replace monomials with different functions such as weighted sums, multi-layer perceptrons and self-attention, thereby streamlining the training of invariant-integration-based architectures. We demonstrate the improved sample complexity on the Rotated-MNIST, SVHN and CIFAR-10 datasets where rotation-invariant-integration-based Wide-ResNet architectures using monomials and weighted sums outperform the respective baselines in the limited sample regime. We achieve state-of-the-art results using full data on Rotated-MNIST and SVHN where rotation is a main source of intraclass variation. On STL-10 we outperform a standard and a rotation-equivariant convolutional neural network using pooling.

cs.LG↗