Jacek Komorowski

Jacek Komorowski

The paper presents a simple and effective learning-based method for computing a discriminative 3D point cloud descriptor for place recognition purposes. Recent state-of-the-art methods have relatively complex architectures such as multi-scale oyramid of point Transformers combined with a pyramid of feature aggregation modules. Our method uses a simple and efficient 3D convolutional feature extraction, based on a sparse voxelized representation, enhanced with channel attention blocks. We employ recent advances in image retrieval and propose a modified version of a loss function based on a differentiable average precision approximation. Such loss function requires training with very large batches for the best results. This is enabled by using multistaged backpropagation. Experimental evaluation on the popular benchmarks proves the effectiveness of our approach, with a consistent improvement over the state of the art

Jacek Komorowski, Grzegorz Kurzejamski

The paper presents a multi-camera tracking method intended for tracking soccer players in long shot video recordings from multiple calibrated cameras installed around the playing field. The large distance to the camera makes it difficult to visually distinguish individual players, which adversely affects the performance of traditional solutions relying on the appearance of tracked objects. Our method focuses on individual player dynamics and interactions between neighborhood players to improve tracking performance. To overcome the difficulty of reliably merging detections from multiple cameras in the presence of calibration errors, we propose the novel tracking approach, where the tracker operates directly on raw detection heat maps from multiple cameras. Our model is trained on a large synthetic dataset generated using Google Research Football Environment and fine-tuned using real-world data to reduce costs involved with ground truth preparation.

Jacek Komorowski, Monika Wysoczanska, Tomasz Trzcinski

We introduce a discriminative multimodal descriptor based on a pair of sensor readings: a point cloud from a LiDAR and an image from an RGB camera. Our descriptor, named MinkLoc++, can be used for place recognition, re-localization and loop closure purposes in robotics or autonomous vehicles applications. We use late fusion approach, where each modality is processed separately and fused in the final part of the processing pipeline. The proposed method achieves state-of-the-art performance on standard place recognition benchmarks. We also identify dominating modality problem when training a multimodal descriptor. The problem manifests itself when the network focuses on a modality with a larger overfit to the training data. This drives the loss down during the training but leads to suboptimal performance on the evaluation set. In this work we describe how to detect and mitigate such risk when using a deep metric learning approach to train a multimodal neural network. Our code is publicly available on the project website: https://github.com/jac99/MinkLocMultimodal.

Jacek Komorowski

The paper presents a learning-based method for computing a discriminative 3D point cloud descriptor for place recognition purposes. Existing methods, such as PointNetVLAD, are based on unordered point cloud representation. They use PointNet as the first processing step to extract local features, which are later aggregated into a global descriptor. The PointNet architecture is not well suited to capture local geometric structures. Thus, state-of-the-art methods enhance vanilla PointNet architecture by adding different mechanism to capture local contextual information, such as graph convolutional networks or using hand-crafted features. We present an alternative approach, dubbed MinkLoc3D, to compute a discriminative 3D point cloud descriptor, based on a sparse voxelized point cloud representation and sparse 3D convolutions. The proposed method has a simple and efficient architecture. Evaluation on standard benchmarks proves that MinkLoc3D outperforms current state-of-the-art. Our code is publicly available on the project website: https://github.com/jac99/MinkLoc3D

Jacek Komorowski, Grzegorz Kurzejamski, Grzegorz Sarwas

The paper describes a deep neural network-based detector dedicated for ball and players detection in high resolution, long shot, video recordings of soccer matches. The detector, dubbed FootAndBall, has an efficient fully convolutional architecture and can operate on input video stream with an arbitrary resolution. It produces ball confidence map encoding the position of the detected ball, player confidence map and player bounding boxes tensor encoding players' positions and bounding boxes. The network uses Feature Pyramid Network desing pattern, where lower level features with higher spatial resolution are combined with higher level features with bigger receptive field. This improves discriminability of small objects (the ball) as larger visual context around the object of interest is taken into account for the classification. Due to its specialized design, the network has two orders of magnitude less parameters than a generic deep neural network-based object detector, such as SSD or YOLO. This allows real-time processing of high resolution input video stream. Our code and pre-trained model can be found on the project website: https://github.com/jac99/FootAndBall .

Kamil Żywanowski, Adam Banaszczyk, Michał R. Nowicki et al.

The 3D LiDAR place recognition aims to estimate a coarse localization in a previously seen environment based on a single scan from a rotating 3D LiDAR sensor. The existing solutions to this problem include hand-crafted point cloud descriptors (e.g., ScanContext, M2DP, LiDAR IRIS) and deep learning-based solutions (e.g., PointNetVLAD, PCAN, LPDNet, DAGC, MinkLoc3D), which are often only evaluated on accumulated 2D scans from the Oxford RobotCar dataset. We introduce MinkLoc3D-SI, a sparse convolution-based solution that utilizes spherical coordinates of 3D points and processes the intensity of 3D LiDAR measurements, improving the performance when a single 3D LiDAR scan is used. Our method integrates the improvements typical for hand-crafted descriptors (like ScanContext) with the most efficient 3D sparse convolutions (MinkLoc3D). Our experiments show improved results on single scans from 3D LiDARs (USyd Campus dataset) and great generalization ability (KITTI dataset). Using intensity information on accumulated 2D scans (RobotCar Intensity dataset) improves the performance, even though spherical representation doesn't produce a noticeable improvement. As a result, MinkLoc3D-SI is suited for single scans obtained from a 3D LiDAR, making it applicable in autonomous vehicles.

Jacek Komorowski, Monika Wysoczanska, Tomasz Trzcinski

The paper presents a deep neural network-based method for global and local descriptors extraction from a point cloud acquired by a rotating 3D LiDAR. The descriptors can be used for two-stage 6DoF relocalization. First, a course position is retrieved by finding candidates with the closest global descriptor in the database of geo-tagged point clouds. Then, the 6DoF pose between a query point cloud and a database point cloud is estimated by matching local descriptors and using a robust estimator such as RANSAC. Our method has a simple, fully convolutional architecture based on a sparse voxelized representation. It can efficiently extract a global descriptor and a set of keypoints with local descriptors from large point clouds with tens of thousand points. Our code and pretrained models are publicly available on the project website.

Jacek Komorowski, Monika Wysoczanska, Tomasz Trzcinski

In this work, we propose a method for large-scale topological localization based on radar scan images using learned descriptors. We present a simple yet efficient deep network architecture to compute a rotationally invariant discriminative global descriptor from a radar scan image. The performance and generalization ability of the proposed method is experimentally evaluated on two large scale driving datasets: MulRan and Oxford Radar RobotCar. Additionally, we present a comparative evaluation of radar-based and LiDAR-based localization using learned global descriptors. Our code and trained models are publicly available on the project website.

Grzegorz Kurzejamski, Jacek Komorowski, Lukasz Dabala et al.

In this paper, we present a framework for computing dense keypoint correspondences between images under strong scene appearance changes. Traditional methods, based on nearest neighbour search in the feature descriptor space, perform poorly when environmental conditions vary, e.g. when images are taken at different times of the day or seasons. Our method improves finding keypoint correspondences in such difficult conditions. First, we use Neighbourhood Consensus Networks to build spatially consistent matching grid between two images at a coarse scale. Then, we apply Superpoint-like corner detector to achieve pixel-level accuracy. Both parts use features learned with domain adaptation to increase robustness against strong scene appearance variations. The framework has been tested on a RobotCar Seasons dataset, proving large improvement on pose estimation task under challenging environmental conditions.

Jacek Komorowski, Grzegorz Kurzejamski, Grzegorz Sarwas

The paper describes a deep network based object detector specialized for ball detection in long shot videos. Due to its fully convolutional design, the method operates on images of any size and produces \emph{ball confidence map} encoding the position of detected ball. The network uses hypercolumn concept, where feature maps from different hierarchy levels of the deep convolutional network are combined and jointly fed to the convolutional classification layer. This allows boosting the detection accuracy as larger visual context around the object of interest is taken into account. The method achieves state-of-the-art results when tested on publicly available ISSIA-CNR Soccer Dataset.

Jacek Komorowski, Przemyslaw Rokita

This paper presents a method for extrinsic camera calibration (estimation of camera rotation and translation matrices) from a sequence of images. It is assumed camera intrinsic matrix and distortion coefficients are known and fixed during the entire sequence. %This allows to decrease a number of pairs of corresponding keypoints between images needed to estimate epipolar geometry compared to uncalibrated case. Performance of the presented method is evaluated on a number of multi-view stereo test datasets. Presented algorithm can be used as a first stage in a dense stereo reconstruction system.

Jacek Komorowski, Przemyslaw Rokita

This paper presents a face recognition method based on a sequence of images. Face shape is reconstructed from images using a combination of structure-from-motion and multi-view stereo methods. The reconstructed 3D face model is compared against models held in a gallery. The novel element in the presented approach is the fact, that the reconstruction is based only on input images and doesn't require a generic, deformable face model. Experimental verification of the proposed method is also included.

Jacek Komorowski, Przemyslaw Rokita

In this paper a method for camera pose estimation from a sequence of images is presented. The method assumes camera is calibrated (intrinsic parameters are known) which allows to decrease a number of required pairs of corresponding points compared to uncalibrated case. Our algorithm can be used as a first stage in a structure from motion stereo reconstruction system.

Jacek Komorowski, Tomasz Trzcinski

In this paper we present an efficient method for aggregating binary feature descriptors to allow compact representation of 3D scene model in incremental structure-from-motion and SLAM applications. All feature descriptors linked with one 3D scene point or landmark are represented by a single low-dimensional real-valued vector called a \emph{prototype}. The method allows significant reduction of memory required to store and process feature descriptors in large-scale structure-from-motion applications. An efficient approximate nearest neighbours search methods suited for real-valued descriptors, such as FLANN, can be used on the resulting prototypes to speed up matching processed frames.

Tomasz Trzcinski, Jacek Komorowski, Lukasz Dabala et al.

Numerous computer vision applications rely on local feature descriptors, such as SIFT, SURF or FREAK, for image matching. Although their local character makes image matching processes more robust to occlusions, it often leads to geometrically inconsistent keypoint matches that need to be filtered out, e.g. using RANSAC. In this paper we propose a novel, more discriminative, descriptor that includes not only local feature representation, but also information about the geometric layout of neighbouring keypoints. To that end, we use a Siamese architecture that learns a low-dimensional feature embedding of keypoint constellation by maximizing the distances between non-corresponding pairs of matched image patches, while minimizing it for correct matches. The 48-dimensional oating point descriptor that we train is built on top of the state-of-the-art FREAK descriptor achieves significant performance improvement over the competitors on a challenging TUM dataset.

Jacek Komorowski, Konrad Czarnota, Tomasz Trzcinski et al.

In the recent years, a number of novel, deep-learning based, interest point detectors, such as LIFT, DELF, Superpoint or LF-Net was proposed. However there's a lack of a standard benchmark to evaluate suitability of these novel keypoint detectors for real-live applications such as autonomous driving. Traditional benchmarks (e.g. Oxford VGG) are rather limited, as they consist of relatively few images of mostly planar scenes taken in favourable conditions. In this paper we verify if the recent, deep-learning based interest point detectors have the advantage over the traditional, hand-crafted keypoint detectors. To this end, we evaluate stability of a number of hand crafted and recent, learning-based interest point detectors on the street-level view ApolloScape dataset.

Jacek Komorowski, Tomasz Trzcinski

In this paper we evaluate performance of data-dependent hashing methods on binary data. The goal is to find a hashing method that can effectively produce lower dimensional binary representation of 512-bit FREAK descriptors. A representative sample of recent unsupervised, semi-supervised and supervised hashing methods was experimentally evaluated on large datasets of labelled binary FREAK feature descriptors.