Local Matching Research Articles

Guide Local Feature Matching by Overlap Estimation

Local image feature matching under large appearance, viewpoint, and distance changes is challenging yet important. Conventional methods detect and match tentative local features across the whole images, with heuristic consistency checks to guarantee reliable matches. In this paper, we introduce a novel Overlap Estimation method conditioned on image pairs with TRansformer, named OETR, to constrain local feature matching in the commonly visible region. OETR performs overlap estimation in a two step process of feature correlation and then overlap regression. As a preprocessing module, OETR can be plugged into any existing local feature detection and matching pipeline, to mitigate potential view angle or scale variance. Intensive experiments show that OETR can boost state of the art local feature matching performance substantially, especially for image pairs with small shared regions. The code will be publicly available at https://github.com/AbyssGaze/OETR.

IIPA-Net: Joint Illumination-Invariant and Pose-Aligned Feature Learning for Person Reidentification

Person reidentification (re-id) has gained significant progress and aroused great interest in computer vision. However, due to the effect of weak illumination and poor alignment, person re-id is still a challenging task. Many previous works focus on either illumination enhancement methods or pose estimation. However, those methods are difficult to apply in real-world scenarios, which usually contain various interference factors. To improve the performance of re-id, we propose an Illumination-Invariant and Pose-Aligned Network (IIPA-Net). The illumination change is handled by a retinex decompose network, and the pose variation problem is solved by a local feature matching method. Based on the multimodal nature of a person, we propose a part attention module to optimize the global feature. Finally, a data-driven training strategy is proposed to train the proposed architecture effectively. Experiments show that the proposed framework outperforms other state-of-the-art approaches on both normal- and low-light datasets.

Framework of metal insertion in the projection domain for image quality optimization in interventional computed tomography.

Purpose: This work aims to develop a framework to accurately and efficiently simulate metallic objects used during interventional oncology (IO) procedures and their artifacts in computed tomography (CT) images of different body regions. Approach: A metal insertion framework based on an existing lesion insertion tool was developed. Noise and beam hardening models were incorporated into the model and validated by comparing images of real and artificially inserted metallic rods of known material composition and dimensions. The framework was further validated by inserting ablation probes into a water phantom and comparing image appearance to scans of real probes at matching locations in the phantom. Finally, a comprehensive library of metallic probes used in our IO practice was generated and a graphical user interface was built to efficiently insert any number of probes at arbitrary positions in patient CT data, including projection and image domain insertions. Results: Metallic rod experiments demonstrated that noise and beam hardening were properly modeled. Phantom and patient data with virtually inserted probes demonstrated similar artifact appearance and magnitude compared with real probes. The developed user interface resulted in accurately co-registered virtual probes both with and without accompanying artifacts from projection and image domain insertions, respectively. Conclusions: The developed metal insertion framework successfully replicates metallic object and artifact appearance with projection domain insertions and provides corresponding artifact-free images with the metallic object in the identical location through image domain insertion. This framework has potential to generate robust training libraries for deep learning algorithms and facilitate image quality optimization in interventional CT.

Abstract 1208: Denoising spatial expression profiling data based on in situ position and image information

Abstract Spatial expression profiling (SEP) technologies provide gene expression close to or even superior to single-cell resolution while retaining the physical locations of sequencing and often also providing matched pathology images. However, the expression data captured by SEP technologies suffer from high noise levels, including but not limited to dropouts as in regular single-cell RNA-sequencing (scRNA-seq). The extra experimental steps for preserving the spatial locations of sequencing could result in even more severe noises, compared to regular scRNA-seq. Fortunately, such noises could be largely reduced by leveraging information from the physical locations of sequencing and the tissue and cellular organization reflected by corresponding pathology images. In this work, we demonstrated the extensive levels of noise in SEP data. We developed a mathematical model, named Sprod, to reduce such noises based on latent space and graph learning of matched location and imaging data. We comprehensively validated Sprod and demonstrated its advantages over prior methods for removing dropouts in scRNA-seq data. We further showed that, after adequately de-noising by Sprod, differential expression analyses, pseudotime analyses, and cell-to-cell interaction inferences yield significantly more informative results in various biological application settings. In particular, with Sprod, we discovered 3-4 times more RNA transcripts that were actively transported in mouse hippocampus neurons. We also showed that the tumor cells at the tumor-stroma boundaries demonstrate differential transcriptomic features from the tumor cells in the central regions, caused by their interactions with the stroma/immune cells. Overall, we envision denoising by Sprod to become a key first step to empower SEP technologies for biomedical discoveries and innovations. Citation Format: Yunguan Wang, Bing Song, Shidan Wang, Mengyi Chen, Yang Xie, Guanghua Xiao, Li Wang, Tao Wang. Denoising spatial expression profiling data based on in situ position and image information [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 1208.

Exploring Match Space: How Medical School and Specialty Characteristics Affect Residency Match Geography in the United States

To explore how school and specialty characteristics impact the geographic match location of U.S. senior medical students. The authors collected student match data between 2018 and 2020 from U.S. MD-granting medical schools and calculated the distance between students' medical schools and residency training programs. They use the term "match space" to describe this distance. Match space was codified on a 5-point ordinal scale by where the student matched: 1 = home institution, 2 = home state, 3 = an adjacent state, 4 = the same or adjacent U.S. Census division (and not adjacent state), and 5 = skipped at least one U.S. Census division. Ordinal logistic regression correlated school and specialty characteristics with match space. During the study period, 26,102 medical students, representing 66 medical schools from 28 states, matched in 23 specialties. Fifty-nine percent of students were from public institutions, and 27% of schools ranked in the top 40 of National Institutes of Health (NIH) research funding. The match space was higher for students graduating from private institutions (odds ratio [OR] 1.14; 95% confidence interval [CI], 1.06 to 1.22) and matching into more competitive specialties (OR 1.07; 95% CI, 1 to 1.14). The match space was lower for students graduating from top NIH-funded institutions (OR 0.89; 95% CI, 0.85 to 0.94) and from schools with a higher percentage of in-state matriculants (OR 0.75; 95% CI, 0.72 to 0.77). School characteristics such as region, public/private designation, NIH funding, and percentage of in-state students were associated with residency match geography. Matching into more competitive specialties also showed a marginal increase in match distance. These findings suggest that a student's choice of specialty and medical school may impact subsequent geographic placement for residency training, which should be considered by students and residency programs alike.

MODL-08. Cellular yield and xenograft creation from probe washing after stereotactic biopsy of deep-seated CNS tumors in children

Abstract Intracranial tumors located in otherwise difficult-to-resect regions of the brain are routinely biopsied to obtain tissue diagnosis and guide clinical decision making. However, tissue yield from these stereotactic biopsy samples is limited and leaves little remaining for research purposes or tumor banking. Given the importance of molecular diagnostics and rise of targeted therapies, research into the genetic and immune landscape of these lesions is of paramount importance, especially for inoperable and deep-seated tumors. We sought to obtain viable cellular material from otherwise discarded biopsy probes without visible tumor sample within the probe. Cells were obtained from clinically challenging pediatric tumors of the brainstem, thalamus, and pineal region by washing biopsy probes in FBS media immediately after completion of a stereotactic biopsy. The wash was then filtered and re-suspended in media before evenly distributing in a 24 well plate for culture purposes. A total of 15 samples were collected from probe washes, including 6 diffuse midline gliomas of the brain stem, 7 supratentorial diffuse midline gliomas, and 2 tumors in the pineal region. Viable tumor cells were obtained from 9/15 samples, ranging from 500 to approximately 525,000 cells per sample wash (219,222 ± 181,951). Orthotopic implantation into matching locations was completed with 7 tumor samples. Formation of a patient-derived orthotopic xenograft was confirmed in one thalamic high-grade glioma (165,000 cells). Three of the five diffuse midline gliomas of the brainstem showed early signs of tumor formation (yielding >2 million cells) and will be validated. In conclusion, viable tumor cells can be collected from biopsy probe washes after stereotactic biopsy, without visible tumor within the probe. Tumor cells can be used for patient-derived orthotopic xenograft tumor formation, which allows for new avenues in the development of animal models of difficult to resect brain tumors.

MODL-29. Molecular Landscape of a comprehensive panel of pediatric brain cancer Patient-derived orthotopic xenograft (PDOX) models inform unique targets for drug responsiveness

Abstract Brain tumor is a leading cause of cancer related death in children. In addition to replicating histopathology, animal models faithfully replicating genetic/epigenetic abnormalities, molecular subtypes and broad inter-tumoral heterogeneities are needed. Through direct implantation of patient surgical or autopsied tumor tissues into matching locations in the brains of SCID mice, we developed a panel of 150 PDOX mouse models. Here, we report the analysis of 74 of the 150 PDOX models, 45 matching patient tissues and 60 non-tumorigenic samples to a well-annotated reference cohort of 2,801 methylation profiles of primary brain tumors. Our data showed that the lack of tumorigenicity was neither correlated with molecular subtypes nor predicted by low cell viabilities of the patient samples. Methylation profiling identified PDOX models representing nearly a full spectrum of molecular subtypes of pediatric brain tumors including GBM, medulloblastoma, ependymoma and ATRT. Direct comparison with the original patient tumors confirmed the replication of molecular subtypes. ONCOplot [FB1] analysis of PDOX models derived from matching pairs of primary and recurrent tumors (n=8) revealed close clustering with the patient tumors. Investigation of metastatic properties was performed in 13 MB models by harvesting and sub-transplanting matching PDOX primary tumors in the cerebella and metastatic tumors in the spinal cords. To confirm the potential and power of PDOX models in preclinical drug testing, we applied fractionated radiation (2 Gy/day x 5 days) and optimized multi-agent combinatory chemotherapies in MB models of the four major subgroups. High-throughput combination drug screening with ~ 8,000 drugs in PDOX-derived GBM cell lines and primary cultures of MB PDOX cells identified a library of ~ 3,500 drugs that were active in pediatric brain tumors. In summary, this study provides detailed information on molecular subclassification of a uniquely large cohort of PDOX models to serve as essential tools for brain tumor research.

Analysis of the Associations between Contextual Variables and Match Running Performance in Croatian First Division Soccer

It has been hypothesized that contextual variables (i.e. match location, opponents’ level) could be important deter- minants of running performance (RP) in soccer matches, but empirical studies provided inconsistent evidences. This study aimed to investigate the associations between contextual variables (CV) and RP in Croatian soccer players. Players’ RP (n=193) were evaluated, and used as cases in this study. CV included match outcome (win-draw-loss), opponent’s level (higher vs. lower ranked), and match location (home-away). RP were measured by global position- ing system during the soccer matches in seasons 2018/19 (n=14) and 2019/20 (n=15) of Croatian first division, and included: the total distance covered, low-intensity (<14.3 km/h), running (14.4–19.7 km/h), high-intensity running (>19.8 km/h), high-speed running (19.8–25.1 km/h) and sprinting (>25.2 km/h). Results indicated significantly great- er amount of distance covered in running zone for away matches (F-test: 7.83, p<0.01, small effect size; 1392±353 m and 1262±294 m, for home and away matches, respectively). Similar RP were observed irrespective of opponents’ level (F-test: 0.05 to 1.53, p>0.05). Lowest total-, low- and running-zone distances were evidenced in won matches (9893±896 m, 8035±614 m and 1241±312 m, respectively), followed by drawn matches (10298±913 m, 8287±838 m and 1363±320 m, respectively) and lost matches (10355±1052 m, 8279±742 m and 1406±343 m, respectively). This study demonstrated that (i) won matches were characterized by lower RP, (ii) association between RP and match location was limited, (iii) RP was not related to the opponents’ level.

THE RELATIONSHIP BETWEEN THE MATCH ANALYSIS RESULTS AND RANKING OF SUCCESS IN TURKISH SPOR TOTO FOOTBALL SUPER LEAGUE

The purpose of the research was to examine the relationship between ranking of success and match analysis parameters, and to develop a regression formula to understand the effectiveness of these parameters on league score in the season of 2018- 2019 Turkish Spor Toto Football Super League. A total of 306 competitions in the league consisting of 18 teams in the season of 2018-2019 were examined. The parameters used in the current research were total, high intensity, and sprint running distances with or without ball, number of successful shots and passes, ball possessions in the first, second and third regions, and points earned at the end of the season. In accordance with the findings, the most related parameters to the league score were total running distance with ball, and the number of successful shots and passes p<0.01) and high intensity running distance with or without ball, sprint running distance, successful passes, and shots were mostly seen in the third region in the season of 2018-2019 Turkish Spor Toto Football Super League. The multivariate linear regression analysis F(7,10)=6.05, p=0.0058 was found statistically significant, and showed that 68 % of the variance in the dependent variable (R2 adjusted=0.68) was explained by the ratio of high intensity running distance with ball to high intensity running distance, sprint distance with ball to sprint distance, and the number successful shots and passes. Future studies may also examine the match location, defensive parameters, opposing teams, and other match analysis parameters.

Influence of the situational variables on the performance of the teams competing in the Chinese Super League

ABSTRACT The aim of this study was to assess the impact of situational variables, match location, quality of the opponent, and final result on the performance of the teams during an entire season (2016) of the Chinese Super League throughout 474 matches. Results showed that playing as an away team reduced the number of positional attacks per game, total ball possession time, possession in the final third of the field, possession in opponent half, passes attempted, passes accurate, crosses, passes forward, picking up free balls, and picking up free balls in opponent half. Final result only affected total attacks made per game and lost ball by the teams that finished the season between the 8th and the 16th position. These teams made an average of 99.29 attacks per game, which was lowered by 1.53 attacks for each goal that the team had in comparison with the opponent at the end of the match. In addition, number of lost balls was also reduced by 1.49 per game for each goal that one team had in comparison with the opponent at the end of the match. Overall, combinations of these variables could be used in order to predict the performance of soccer teams.

The case of electoral polling station data for geocoding in facilitating accessibility to social, economic and cultural opportunities in Ghana

ABSTRACT It is widely acknowledged that geographic data in Ghana and most of Sub-Sahara Africa is patchy, and lack geographic coverage and detail. This has implications for mobility and access to socioeconomic, and cultural opportunities which require address and location data typically provided through geocoding. A critical component of geocoding is a reference database to lookup matching address and location information. Accordingly, we investigate the suitability of, and the use of electoral polling stations to build a reference database for geocoding. This was undertaken by mapping their geographic coordinates, analyzing their geographic spread and detail, and outlining the database structure for building the reference database. Using the 2020 electoral polling station data for the Greater North area of Ghana, we found that 93% of the geographic coordinates can be obtained from open data sources. Also, there is high proximity with a nearest neighbor distance of 663 m, a reasonable service area of 4300 m, and a median distance of 3700 m from a community location. The widespread availability, large geographic spread and moderate detail, make polling stations a creditable data source for geocoding. It is recommended the polling station data be augmented with a detailed geographic data like those from landmarks and paratransit service.

Learning reliable modal weight with transformer for robust RGBT tracking

Many Siamese-based RGBT trackers have been prevalently designed in recent years for fast-tracking. However, the correlation operation in them is a local linear matching process, which may easily lose semantic information required inevitably by high-precision trackers. In this paper, we propose a strong cross-modal model based on transformer for robust RGBT tracking. Specifically, a simple dual-flow convolutional network is designed to extract and fuse dual-modal features, with comparably lower complexity. Besides, to enhance the feature representation and deepen semantic features, a modal weight allocation strategy and a backbone feature extracted network based on modified Resnet-50 are designed, respectively. Also, an attention-based transformer feature fusion network is adopted to improve long-distance feature association to decrease the loss of semantic information. Finally, a classification regression subnetwork is investigated to accurately predict the state of the target. Sufficient experiments have been implemented on the RGBT234, RGBT210, GTOT and LasHeR datasets, demonstrating more outstanding tracking performance against the state-of-the-art RGBT trackers.

Local spatial alignment network for few-shot learning

Few-Shot Learning (FSL) aims at recognizing the novel classes with extremely limited samples via transferring the learned knowledge from some base classes. Most of the existing metric-based approaches focus on measuring the instance-level feature similarity but neglect the spatial alignment between different instances, which would lead to poor adaptation with inaccurate local spatial matching, especially for he fine-grained classes. In this paper, we propose a model called Local Spatial Alignment Network (LSANet) to measure the instance-to-class similarity via aligning the local spatial regions in a traversal scanning way. Specifically, the local spatial alignment is achieved by continuously sampling the local patches from the query feature map, where each local patch performs as a kernel to filter the most similar local patches from the support feature maps, obtaining the patch-level similarities between the query instance and the support classes. Then, we propose an information aggregation module to aggregate the patch-level similarities into the class prediction score, where the important patches are highlighted and the backgrounds are diluted. In this way, our model is able to both align the local spatial patches and capture the discriminative information, which benefits in adapting for the novel few-shot classes. To evaluate the effectiveness of the proposed model, we conduct extensive experiments on both coarse-grained and fine-grained datasets. The experimental results show that the proposed LSANet performs competitively on the FSL benchmarks of different granularities.

Global and local interaction matching model for knowledge-grounded response selection in retrieval-based chatbots

Knowledge-grounded response selection, the task of selecting a proper response from a set of response candidates given the conversation context and its background knowledge, is important for building intelligent dialogue systems. Challenges for the task mainly lie in how to interact between the context and the knowledge, and how to match between the response and the context-knowledge pair. Existing work mainly focuses on exploiting global or local interaction matching information, and hence cannot fully utilize the information. In this paper, we propose a global and local interaction matching model (GLIMM) that matches a response with the context-knowledge pair from global and local views. For global interaction matching, GLIMM takes the context and the knowledge as long sequences. For local interaction matching, GLIMM treats each context utterance and knowledge sentence separately. GLIMM integrates information from two different views that could provide complementary information for each other. Experiments on two datasets show that the proposed model outperforms the state-of-the-art baselines.

Implementation of a Human Motion Capture System Based on the Internet of Things Machine Vision

The classification of the stereo matching comprehensive analysis related algorithm model can be subdivided into local stereo matching based on the entire acquisition and global stereo matching based on the entire local. But it can have a higher capture efficiency because the log-likelihood variance cost calculation function can have a faster feature convergence capture speed than the ordinary log-mean-square error cost function. Through the combination of gray channel and frame difference channel, a better network structure and parameters on the KTH data set are obtained, which can ensure the classification effect while greatly reducing the number of parameters, improving training efficiency and improving classification accuracy. The article uses dual-channel 3D convolutional human neural network technology to achieve 92.5% accuracy of human feature capture, which is significantly better than many traditional feature extraction techniques proposed in the literature.

Postmatch Recovery Practices Carried Out in Professional Football: A Survey of 56 Portuguese Professional Football Teams.

In football (ie,soccer), the effectiveness of recovery practices is critical to maintain high levels of performance, prevent residual fatigue, and reduce injury risk. However, the recovery methods that professional football teams put in place up to 72hours postmatch are widely unknown. This study aimed to characterize the postmatch recovery practices in Portuguese football teams in different postmatch periods and match locations. Portuguese football teams (total, N = 56; male: first league [n = 17], second league [n = 16], under-23 league [n = 12]; female: first league, n = 11) participated in the study. The practitioners in charge of recovery strategies in each team filled out the online questionnaire in the middle of the 2019-20 season. The questions focused on the type of recovery methods to be used at different times after home and away matches. After home matches, stretching, electrostimulation, active recovery, and massage were used with higher frequency (P < .017) in later postmatch periods (ie,12-24, and 24-72h) compared with immediately postmatch. After away matches, several differences were found (P < .017) for the stretching, electrostimulation, active recovery, cold-water immersion, massage, nutrition, and sleep between postmatch periods. Regarding match location, stretching (r = .19), active recovery (r = .39), cold-water immersion (r = .46), and massage (r = .29) showed a higher frequency of use immediately after home matches. Conversely, the use of compression garments (r = .27) was higher immediately after away matches. It was concluded that in professional football, recovery methods are not applied uniformly along postmatch periods and differ depending on the match location.

Stereo slant discrimination of planar 3D surfaces: Frontoparallel versus planar matching.

Binocular stereo cues are important for discriminating 3D surface orientation, especially at near distances. We devised a single-interval task where observers discriminated the slant of a densely textured planar test surface relative to a textured planar surround reference surface. Although surfaces were rendered with correct perspective, the stimuli were designed so that the binocular cues dominated performance. Slant discrimination performance was measured as a function of the reference slant and the level of uncorrelated white noise added to the test-plane images in the left and right eyes. We compared human performance with an approximate ideal observer (planar matching [PM]) and two subideal observers. The PM observer uses the image in one eye and back projection to predict a test image in the other eye for all possible slants, tilts, and distances. The estimated slant, tilt, and distance are determined by the prediction that most closely matches the measured image in the other eye. The first subideal observer (local planar matching [LPM]) applies PM over local neighborhoods and then pools estimates across the test plane. The second suboptimal observer (local frontoparallel matching [LFM]) uses only location disparity. We find that the ideal observer (PM) and the first subideal observer (LPM) outperforms the second subideal observer (LFM), demonstrating the additional benefit of pattern disparities. We also find that all three model observers can account for human performance, if two free parameters are included: a fixed small level of internal estimation noise, and a fixed overall efficiency scalar on slant discriminability.

Effect of match-related contextual factors on positional performance in the national rugby league

Objectives To examine the effects of match-related contextual variables on positional groups and success in the National Rugby League (NRL). Methods Data relating to match location, match outcome, quality of opposition and match type (absolute score differential) from all matches across the 2015–2019 NRL seasons were collected, in addition to 14 previously identified Factors (technical performance indicators). A decision tree, grown using the Exhaustive Chi-square Automatic Interaction Detector (CHAID) algorithm, was used to model the effect of each of these match-related contexts on positional contribution according to match outcome. Results The accuracy of the exhaustive CHAID model in explaining the influence of positional groups on match outcome was 66%. The model revealed four primary splits: interchange forwards, utility backs, adjustables and a group containing the remaining three positional groups (forwards, backs, and interchange). Conclusions Results suggest that interchange forwards, utility backs and adjustables could have a definitive role within the team compared to the remaining positional groups in determining match outcome. In contrast to team-level research, there is a greater emphasis on the importance of defensive actions (e.g. try causes, tackles made) at a positional level than attacking performance indicators. The moderate classification accuracy justifies the use of this approach for examination of the interactions between match-related contextual variables, performance indicators and positional groups.

Radargrammetric DSM Generation by Semi-Global Matching and Evaluation of Penalty Functions

Radargrammetry is a useful approach to generate Digital Surface Models (DSMs) and an alternative to InSAR techniques that are subject to temporal or atmospheric decorrelation. Stereo image matching in radargrammetry refers to the process of determining homologous points in two images. The performance of image matching influences the final quality of DSM used for spatial-temporal analysis of landscapes and terrain. In SAR image matching, local matching methods are commonly used but usually produce sparse and inaccurate homologous points adding ambiguity to final products; global or semi-global matching methods are seldom applied even though more accurate and dense homologous points can be yielded. To fill this gap, we propose a hierarchical semi-global matching (SGM) pipeline to reconstruct DSMs in forested and mountainous regions using stereo TerraSAR-X images. In addition, three penalty functions were implemented in the pipeline and evaluated for effectiveness. To make accuracy and efficiency comparisons between our SGM dense matching method and the local matching method, the normalized cross-correlation (NCC) local matching method was also applied to generate DSMs using the same test data. The accuracy of radargrammetric DSMs was validated against an airborne photogrammetric reference DSM and compared with the accuracy of NASA’s 30 m SRTM DEM. The results show the SGM pipeline produces DSMs with height accuracy and computing efficiency that exceeds the SRTM DEM and NCC-derived DSMs. The penalty function adopting the Canny edge detector yields a higher vertical precision than the other two evaluated penalty functions. SGM is a powerful and efficient tool to produce high-quality DSMs using stereo Spaceborne SAR images.

Loop-migration and non-breeding locations of British breeding Wood Warblers Phylloscopus sibilatrix

ABSTRACT Capsule: British breeding Wood Warblers Phylloscopus sibilatrix show a clockwise loop migration incorporating stops in southern Europe, the Sahel, and the humid forest zone of West Africa. Aims: To determine autumn and spring migration routes, the location and duration of stopover sites on migration, and the location of non-breeding areas of British breeding Wood Warblers. Methods: In 2016 and 2018 we deployed geolocators to male Wood Warblers on Dartmoor, Devon, and in the New Forest, Hampshire. We retrieved four geolocators from returning birds in 2017, 2019, and 2020. Results: Male Wood Warblers departed breeding sites in late July and stopped for most of August in central southern Europe, crossed the Sahara by a non-stop night and day flight immediately followed by a short stop, and then migrated west to a longer stopover in the Sahel. Final non-breeding destinations were in an area of West Africa covering Guinea, Sierra Leone, and Liberia. Two were tracked on spring migration, again crossing the Sahara via a non-stop flight before migrating through Western Europe to complete a clockwise loop migration back to Britain. Conclusion: All tracked Wood Warblers used stopovers for at least three weeks in three distinct regions, in central southern Europe, in the Sahel, and in the humid zone of West Africa. Although the limitations of geolocation prevent matching locations with habitat, these regions are broadly characterized by distinct forest or woodland habitat types, which differ from breeding habitat. All four tracks showed similar patterns in route, stopover behaviour, and timings, suggesting they may be representative of males in these breeding populations, and potentially of other British and western European Wood Warbler populations.