Iaroslav Melekhov

Jyri Maanpää, Iaroslav Melekhov, Josef Taher et al.

Robustness of different pattern recognition methods is one of the key challenges in autonomous driving, especially when driving in the high variety of road environments and weather conditions, such as gravel roads and snowfall. Although one can collect data from these adverse conditions using cars equipped with sensors, it is quite tedious to annotate the data for training. In this work, we address this limitation and propose a CNN-based method that can leverage the steering wheel angle information to improve the road area semantic segmentation. As the steering wheel angle data can be easily acquired with the associated images, one could improve the accuracy of road area semantic segmentation by collecting data in new road environments without manual data annotation. We demonstrate the effectiveness of the proposed approach on two challenging data sets for autonomous driving and show that when the steering task is used in our segmentation model training, it leads to a 0.1-2.9% gain in the road area mIoU (mean Intersection over Union) compared to the corresponding reference transfer learning model.

Zakaria Laskar, Iaroslav Melekhov, Assia Benbihi et al.

Camera relocalization relies on 3D models of the scene with a large memory footprint that is incompatible with the memory budget of several applications. One solution to reduce the scene memory size is map compression by removing certain 3D points and descriptor quantization. This achieves high compression but leads to performance drop due to information loss. To address the memory performance trade-off, we train a light-weight scene-specific auto-encoder network that performs descriptor quantization-dequantization in an end-to-end differentiable manner updating both product quantization centroids and network parameters through back-propagation. In addition to optimizing the network for descriptor reconstruction, we encourage it to preserve the descriptor-matching performance with margin-based metric loss functions. Results show that for a local descriptor memory of only 1MB, the synergistic combination of the proposed network and map compression achieves the best performance on the Aachen Day-Night compared to existing compression methods.

Iira Häkkinen, Iaroslav Melekhov, Erik Englesson et al.

The field of medical image segmentation is hindered by the scarcity of large, publicly available annotated datasets. Not all datasets are made public for privacy reasons, and creating annotations for a large dataset is time-consuming and expensive, as it requires specialized expertise to accurately identify regions of interest (ROIs) within the images. To address these challenges, we evaluate the performance of the Segment Anything Model (SAM) as an annotation tool for medical data by using it to produce so-called "pseudo labels" on the Medical Segmentation Decathlon (MSD) computed tomography (CT) tasks. The pseudo labels are then used in place of ground truth labels to train a UNet model in a weakly-supervised manner. We experiment with different prompt types on SAM and find that the bounding box prompt is a simple yet effective method for generating pseudo labels. This method allows us to develop a weakly-supervised model that performs comparably to a fully supervised model.

Xuqian Ren, Matias Turkulainen, Jiepeng Wang et al.

Geometric priors are often used to enhance 3D reconstruction. With many smartphones featuring low-resolution depth sensors and the prevalence of off-the-shelf monocular geometry estimators, incorporating geometric priors as regularization signals has become common in 3D vision tasks. However, the accuracy of depth estimates from mobile devices is typically poor for highly detailed geometry, and monocular estimators often suffer from poor multi-view consistency and precision. In this work, we propose an approach for joint surface depth and normal refinement of Gaussian Splatting methods for accurate 3D reconstruction of indoor scenes. We develop supervision strategies that adaptively filters low-quality depth and normal estimates by comparing the consistency of the priors during optimization. We mitigate regularization in regions where prior estimates have high uncertainty or ambiguities. Our filtering strategy and optimization design demonstrate significant improvements in both mesh estimation and novel-view synthesis for both 3D and 2D Gaussian Splatting-based methods on challenging indoor room datasets. Furthermore, we explore the use of alternative meshing strategies for finer geometry extraction. We develop a scale-aware meshing strategy inspired by TSDF and octree-based isosurface extraction, which recovers finer details from Gaussian models compared to other commonly used open-source meshing tools. Our code is released in https://xuqianren.github.io/ags_mesh_website/.

Matias Turkulainen, Xuqian Ren, Iaroslav Melekhov et al.

High-fidelity 3D reconstruction of common indoor scenes is crucial for VR and AR applications. 3D Gaussian splatting, a novel differentiable rendering technique, has achieved state-of-the-art novel view synthesis results with high rendering speeds and relatively low training times. However, its performance on scenes commonly seen in indoor datasets is poor due to the lack of geometric constraints during optimization. In this work, we explore the use of readily accessible geometric cues to enhance Gaussian splatting optimization in challenging, ill-posed, and textureless scenes. We extend 3D Gaussian splatting with depth and normal cues to tackle challenging indoor datasets and showcase techniques for efficient mesh extraction. Specifically, we regularize the optimization procedure with depth information, enforce local smoothness of nearby Gaussians, and use off-the-shelf monocular networks to achieve better alignment with the true scene geometry. We propose an adaptive depth loss based on the gradient of color images, improving depth estimation and novel view synthesis results over various baselines. Our simple yet effective regularization technique enables direct mesh extraction from the Gaussian representation, yielding more physically accurate reconstructions of indoor scenes.

Iaroslav Melekhov, Anand Umashankar, Hyeong-Jin Kim et al.

We introduce ECLAIR (Extended Classification of Lidar for AI Recognition), a new outdoor large-scale aerial LiDAR dataset designed specifically for advancing research in point cloud semantic segmentation. As the most extensive and diverse collection of its kind to date, the dataset covers a total area of 10$km^2$ with close to 600 million points and features eleven distinct object categories. To guarantee the dataset's quality and utility, we have thoroughly curated the point labels through an internal team of experts, ensuring accuracy and consistency in semantic labeling. The dataset is engineered to move forward the fields of 3D urban modeling, scene understanding, and utility infrastructure management by presenting new challenges and potential applications. As a benchmark, we report qualitative and quantitative analysis of a voxel-based point cloud segmentation approach based on the Minkowski Engine.

Jyri Maanpää, Julius Pesonen, Heikki Hyyti et al.

Slippery road weather conditions are prevalent in many regions and cause a regular risk for traffic. Still, there has been less research on how autonomous vehicles could detect slippery driving conditions on the road to drive safely. In this work, we propose a method to predict a dense grip map from the area in front of the car, based on postprocessed multimodal sensor data. We trained a convolutional neural network to predict pixelwise grip values from fused RGB camera, thermal camera, and LiDAR reflectance images, based on weakly supervised ground truth from an optical road weather sensor. The experiments show that it is possible to predict dense grip values with good accuracy from the used data modalities as the produced grip map follows both ground truth measurements and local weather conditions, such as snowy areas on the road. The model using only the RGB camera or LiDAR reflectance modality provided good baseline results for grip prediction accuracy while using models fusing the RGB camera, thermal camera, and LiDAR modalities improved the grip predictions significantly.

Junyuan Fang, Zihan Wang, Yejun Zhang et al.

Vision-language models (VLMs) have demonstrated impressive zero-shot transfer capabilities in image-level visual perception tasks. However, they fall short in 3D instance-level segmentation tasks that require accurate localization and recognition of individual objects. To bridge this gap, we introduce a novel 3D Gaussian Splatting based hard visual prompting approach that leverages camera interpolation to generate diverse viewpoints around target objects without any 2D-3D optimization or fine-tuning. Our method simulates realistic 3D perspectives, effectively augmenting existing hard visual prompts by enforcing geometric consistency across viewpoints. This training-free strategy seamlessly integrates with prior hard visual prompts, enriching object-descriptive features and enabling VLMs to achieve more robust and accurate 3D instance segmentation in diverse 3D scenes.

Jyri Maanpää, Julius Pesonen, Iaroslav Melekhov et al.

Slippery road conditions pose significant challenges for autonomous driving. Beyond predicting road grip, it is crucial to estimate its uncertainty reliably to ensure safe vehicle control. In this work, we benchmark several uncertainty prediction methods to assess their effectiveness for grip uncertainty estimation. Additionally, we propose a novel approach that leverages road surface state segmentation to predict grip uncertainty. Our method estimates a pixel-wise grip probability distribution based on inferred road surface conditions. Experimental results indicate that the proposed approach enhances the robustness of grip uncertainty prediction.

Zakaria Laskar, Tomas Vojir, Matej Grcic et al.

Before deployment in the real-world deep neural networks require thorough evaluation of how they handle both knowns, inputs represented in the training data, and unknowns (anomalies). This is especially important for scene understanding tasks with safety critical applications, such as in autonomous driving. Existing datasets allow evaluation of only knowns or unknowns - but not both, which is required to establish "in the wild" suitability of deep neural network models. To bridge this gap, we propose a novel anomaly segmentation dataset, ISSU, that features a diverse set of anomaly inputs from cluttered real-world environments. The dataset is twice larger than existing anomaly segmentation datasets, and provides a training, validation and test set for controlled in-domain evaluation. The test set consists of a static and temporal part, with the latter comprised of videos. The dataset provides annotations for both closed-set (knowns) and anomalies, enabling closed-set and open-set evaluation. The dataset covers diverse conditions, such as domain and cross-sensor shift, illumination variation and allows ablation of anomaly detection methods with respect to these variations. Evaluation results of current state-of-the-art methods confirm the need for improvements especially in domain-generalization, small and large object segmentation.

Shuzhe Wang, Zakaria Laskar, Iaroslav Melekhov et al.

Visual localization is critical to many applications in computer vision and robotics. To address single-image RGB localization, state-of-the-art feature-based methods match local descriptors between a query image and a pre-built 3D model. Recently, deep neural networks have been exploited to regress the mapping between raw pixels and 3D coordinates in the scene, and thus the matching is implicitly performed by the forward pass through the network. However, in a large and ambiguous environment, learning such a regression task directly can be difficult for a single network. In this work, we present a new hierarchical scene coordinate network to predict pixel scene coordinates in a coarse-to-fine manner from a single RGB image. The proposed method, which is an extension of HSCNet, allows us to train compact models which scale robustly to large environments. It sets a new state-of-the-art for single-image localization on the 7-Scenes, 12 Scenes, Cambridge Landmarks datasets, and the combined indoor scenes.

Iaroslav Melekhov, Zakaria Laskar, Xiaotian Li et al.

Fully-supervised CNN-based approaches for learning local image descriptors have shown remarkable results in a wide range of geometric tasks. However, most of them require per-pixel ground-truth keypoint correspondence data which is difficult to acquire at scale. To address this challenge, recent weakly- and self-supervised methods can learn feature descriptors from relative camera poses or using only synthetic rigid transformations such as homographies. In this work, we focus on understanding the limitations of existing self-supervised approaches and propose a set of improvements that combined lead to powerful feature descriptors. We show that increasing the search space from in-pair to in-batch for hard negative mining brings consistent improvement. To enhance the discriminativeness of feature descriptors, we propose a coarse-to-fine method for mining local hard negatives from a wider search space by using global visual image descriptors. We demonstrate that a combination of synthetic homography transformation, color augmentation, and photorealistic image stylization produces useful representations that are viewpoint and illumination invariant. The feature descriptors learned by the proposed approach perform competitively and surpass their fully- and weakly-supervised counterparts on various geometric benchmarks such as image-based localization, sparse feature matching, and image retrieval.

Shuzhe Wang, Zakaria Laskar, Iaroslav Melekhov et al.

For several emerging technologies such as augmented reality, autonomous driving and robotics, visual localization is a critical component. Directly regressing camera pose/3D scene coordinates from the input image using deep neural networks has shown great potential. However, such methods assume a stationary data distribution with all scenes simultaneously available during training. In this paper, we approach the problem of visual localization in a continual learning setup -- whereby the model is trained on scenes in an incremental manner. Our results show that similar to the classification domain, non-stationary data induces catastrophic forgetting in deep networks for visual localization. To address this issue, a strong baseline based on storing and replaying images from a fixed buffer is proposed. Furthermore, we propose a new sampling method based on coverage score (Buff-CS) that adapts the existing sampling strategies in the buffering process to the problem of visual localization. Results demonstrate consistent improvements over standard buffering methods on two challenging datasets -- 7Scenes, 12Scenes, and also 19Scenes by combining the former scenes.

Jyri Maanpää, Josef Taher, Petri Manninen et al.

Autonomous driving is challenging in adverse road and weather conditions in which there might not be lane lines, the road might be covered in snow and the visibility might be poor. We extend the previous work on end-to-end learning for autonomous steering to operate in these adverse real-life conditions with multimodal data. We collected 28 hours of driving data in several road and weather conditions and trained convolutional neural networks to predict the car steering wheel angle from front-facing color camera images and lidar range and reflectance data. We compared the CNN model performances based on the different modalities and our results show that the lidar modality improves the performances of different multimodal sensor-fusion models. We also performed on-road tests with different models and they support this observation.

Iaroslav Melekhov, Gabriel J. Brostow, Juho Kannala et al.

Local features that are robust to both viewpoint and appearance changes are crucial for many computer vision tasks. In this work we investigate if photorealistic image stylization improves robustness of local features to not only day-night, but also weather and season variations. We show that image stylization in addition to color augmentation is a powerful method of learning robust features. We evaluate learned features on visual localization benchmarks, outperforming state of the art baseline models despite training without ground-truth 3D correspondences using synthetic homographies only. We use trained feature networks to compete in Long-Term Visual Localization and Map-based Localization for Autonomous Driving challenges achieving competitive scores.

Roman Solovyev, Iaroslav Melekhov, Timo Lesonen et al.

Cardiothoratic ratio (CTR) estimated from chest radiographs is a marker indicative of cardiomegaly, the presence of which is in the criteria for heart failure diagnosis. Existing methods for automatic assessment of CTR are driven by Deep Learning-based segmentation. However, these techniques produce only point estimates of CTR but clinical decision making typically assumes the uncertainty. In this paper, we propose a novel method for chest X-ray segmentation and CTR assessment in an automatic manner. In contrast to the previous art, we, for the first time, propose to estimate CTR with uncertainty bounds. Our method is based on Deep Convolutional Neural Network with Feature Pyramid Network (FPN) decoder. We propose two modifications of FPN: replace the batch normalization with instance normalization and inject the dropout which allows to obtain the Monte-Carlo estimates of the segmentation maps at test time. Finally, using the predicted segmentation mask samples, we estimate CTR with uncertainty. In our experiments we demonstrate that the proposed method generalizes well to three different test sets. Finally, we make the annotations produced by two radiologists for all our datasets publicly available.

Aleksei Tiulpin, Iaroslav Melekhov, Simo Saarakkala

This paper addresses the challenge of localization of anatomical landmarks in knee X-ray images at different stages of osteoarthritis (OA). Landmark localization can be viewed as regression problem, where the landmark position is directly predicted by using the region of interest or even full-size images leading to large memory footprint, especially in case of high resolution medical images. In this work, we propose an efficient deep neural networks framework with an hourglass architecture utilizing a soft-argmax layer to directly predict normalized coordinates of the landmark points. We provide an extensive evaluation of different regularization techniques and various loss functions to understand their influence on the localization performance. Furthermore, we introduce the concept of transfer learning from low-budget annotations, and experimentally demonstrate that such approach is improving the accuracy of landmark localization. Compared to the prior methods, we validate our model on two datasets that are independent from the train data and assess the performance of the method for different stages of OA severity. The proposed approach demonstrates better generalization performance compared to the current state-of-the-art.

Zakaria Laskar, Iaroslav Melekhov, Hamed R. Tavakoli et al.

This paper addresses the problem of determining dense pixel correspondences between two images and its application to geometric correspondence verification in image retrieval. The main contribution is a geometric correspondence verification approach for re-ranking a shortlist of retrieved database images based on their dense pair-wise matching with the query image at a pixel level. We determine a set of cyclically consistent dense pixel matches between the pair of images and evaluate local similarity of matched pixels using neural network based image descriptors. Final re-ranking is based on a novel similarity function, which fuses the local similarity metric with a global similarity metric and a geometric consistency measure computed for the matched pixels. For dense matching our approach utilizes a modified version of a recently proposed dense geometric correspondence network (DGC-Net), which we also improve by optimizing the architecture. The proposed model and similarity metric compare favourably to the state-of-the-art image retrieval methods. In addition, we apply our method to the problem of long-term visual localization demonstrating promising results and generalization across datasets.

Iaroslav Melekhov, Aleksei Tiulpin, Torsten Sattler et al.

This paper addresses the challenge of dense pixel correspondence estimation between two images. This problem is closely related to optical flow estimation task where ConvNets (CNNs) have recently achieved significant progress. While optical flow methods produce very accurate results for the small pixel translation and limited appearance variation scenarios, they hardly deal with the strong geometric transformations that we consider in this work. In this paper, we propose a coarse-to-fine CNN-based framework that can leverage the advantages of optical flow approaches and extend them to the case of large transformations providing dense and subpixel accurate estimates. It is trained on synthetic transformations and demonstrates very good performance to unseen, realistic, data. Further, we apply our method to the problem of relative camera pose estimation and demonstrate that the model outperforms existing dense approaches.

Iaroslav Melekhov, Juho Kannala, Esa Rahtu

In this work we propose a neural network based image descriptor suitable for image patch matching, which is an important task in many computer vision applications. Our approach is influenced by recent success of deep convolutional neural networks (CNNs) in object detection and classification tasks. We develop a model which maps the raw input patch to a low dimensional feature vector so that the distance between representations is small for similar patches and large otherwise. As a distance metric we utilize L2 norm, i.e. Euclidean distance, which is fast to evaluate and used in most popular hand-crafted descriptors, such as SIFT. According to the results, our approach outperforms state-of-the-art L2-based descriptors and can be considered as a direct replacement of SIFT. In addition, we conducted experiments with batch normalization and histogram equalization as a preprocessing method of the input data. The results confirm that these techniques further improve the performance of the proposed descriptor. Finally, we show promising preliminary results by appending our CNNs with recently proposed spatial transformer networks and provide a visualisation and interpretation of their impact.

Zakaria Laskar, Iaroslav Melekhov, Surya Kalia et al.

We propose a new deep learning based approach for camera relocalization. Our approach localizes a given query image by using a convolutional neural network (CNN) for first retrieving similar database images and then predicting the relative pose between the query and the database images, whose poses are known. The camera location for the query image is obtained via triangulation from two relative translation estimates using a RANSAC based approach. Each relative pose estimate provides a hypothesis for the camera orientation and they are fused in a second RANSAC scheme. The neural network is trained for relative pose estimation in an end-to-end manner using training image pairs. In contrast to previous work, our approach does not require scene-specific training of the network, which improves scalability, and it can also be applied to scenes which are not available during the training of the network. As another main contribution, we release a challenging indoor localisation dataset covering 5 different scenes registered to a common coordinate frame. We evaluate our approach using both our own dataset and the standard 7 Scenes benchmark. The results show that the proposed approach generalizes well to previously unseen scenes and compares favourably to other recent CNN-based methods.

Iaroslav Melekhov, Juha Ylioinas, Juho Kannala et al.

In this paper, we propose an encoder-decoder convolutional neural network (CNN) architecture for estimating camera pose (orientation and location) from a single RGB-image. The architecture has a hourglass shape consisting of a chain of convolution and up-convolution layers followed by a regression part. The up-convolution layers are introduced to preserve the fine-grained information of the input image. Following the common practice, we train our model in end-to-end manner utilizing transfer learning from large scale classification data. The experiments demonstrate the performance of the approach on data exhibiting different lighting conditions, reflections, and motion blur. The results indicate a clear improvement over the previous state-of-the-art even when compared to methods that utilize sequence of test frames instead of a single frame.

Iaroslav Melekhov, Juha Ylioinas, Juho Kannala et al.

This paper presents a convolutional neural network based approach for estimating the relative pose between two cameras. The proposed network takes RGB images from both cameras as input and directly produces the relative rotation and translation as output. The system is trained in an end-to-end manner utilising transfer learning from a large scale classification dataset. The introduced approach is compared with widely used local feature based methods (SURF, ORB) and the results indicate a clear improvement over the baseline. In addition, a variant of the proposed architecture containing a spatial pyramid pooling (SPP) layer is evaluated and shown to further improve the performance.