Our Work

A resource to support recently onboarded member organisations to The Carpentries. 

Temporary Description

Description This repository contains the research data for "CEFFD: A benchmark dataset and evaluation framework for freshwater fish detection in Central European fishways" [under review]. This in…

Description This repository contains the research data for "CEFFD: A benchmark dataset and evaluation framework for freshwater fish detection in Central European fishways" [under review]. This includes the proposed benchmark's dataset (dataset.zip) and evaluation framework (benchmark.zip) for fish detection training and evaluation in Central European fishways. This benchmark can be freely used and extended to support transparent and reliable automated fish detection in fishways. Additionally, all model weights, training scripts, raw and formatted results are provided to support the reproducibility of the study. The dataset contains 85,931 annotated frames and 94,462 fish bounding boxes (no species labels) collected from five Central European rivers: the Main, Moselle, Elde, Aller and Neckar. The training subset contains 58,897 frames dedicated to a five-fold training loop (roughly 47,117 training and 11,780 validation frames per cross-validation). The evaluation subset contains 27,034 frames. The Aller and Neckar rivers are held out for cross-site evaluation (5,942 frames for Aller River and 5,959 frames for Neckar River). Frames originate from 12 video collections, which are used in the benchmark to determine evaluation subsets: 13 14 51 52 29 30 43 44 100 101 2425 1213 Main Main Main Main Elde Elde Elde Elde Moselle Moselle Neckar Aller Therefore, subset 13-101 corresponds to the Main, Moselle and Elde rivers, 1213 corresponds to the Aller River, and 2425 corresponds to the Neckar River. The benchmark has three tasks: fish detection, fish/no-fish video classification and fish instance counting (more details can be seen in README.md in benchmark.zip). Below are fish detection results on all evaluation frames for models used in the study (provided in model_weights.zip): YOLOv5s, YOLOv8s, YOLO11s, YOLO12s and YOLO26s. Model P@thresh R@thresh F1@thresh Threshold P-AUC R-AUC F1-AUC mAP@50 mAP@50-95 YOLOv5s 0.946 0.895 0.920 0.320 0.931 0.777 0.824 0.942 0.728 YOLOv8s 0.950 0.889 0.919 0.322 0.923 0.751 0.799 0.937 0.734 YOLO11s 0.950 0.881 0.914 0.324 0.922 0.741 0.793 0.928 0.726 YOLO12s 0.951 0.886 0.918 0.289 0.925 0.736 0.789 0.933 0.732 YOLO26s 0.938 0.850 0.892 0.219 0.921 0.696 0.765 0.906 0.706 Content dataset_sample.zip - Contains a small sample (preview) of the dataset to check the content and the structure before downloading the dataset.zip. dataset.zip - Full dataset containing dataset.json meta-file and images/ folder with frames. benchmark.zip - Contains evaluation scripts (evaluate.py and analyse_results.py) for model evaluation. The benchmark is executed from the command line. An example notebook is provided (also used in the study) to automate the evaluation of all models using the benchmark. model_weights.zip - Contains 25 model files (five per architecture) as well as the training scripts these models were trained with. raw_results.zip - Contains the results of running the evaluate.py over all models (contains also extra results, such as confidence-curve axis values). formatted_results.zip - Results presented within the study. Obtained by running analyse_results.py over raw results. How to use To use the benchmark, download both dataset.zip and benchmark.zip. Custom datasets can be made, following the same dataset structure. Similarly, the benchmark can be extended for other models (not restricted to the YOLO family) following the README.md. To get the results, either download formatted_results.zip or download benchmark.zip and raw_results.zip and run the analyse_results.py script. To reproduce the results,download both dataset.zip and benchmark.zip. Use the convert_dataset.py scripts to convert it into a training format (supported YOLO and COCO). In this study, the YOLO format was used for training. Use the provided training script to train the models. The conversion script will automatically create five cross-validation subsets; therefore, there will be 25 trained models. After training the models, use the last models’ weights with a corresponding model wrapper in benchmark (yolov5 for YOLOv5 models and ultralytics for YOLOv8, YOLO11, YOLO12 and YOLO26). Run the benchmark, which will produce 25 result folders, one for each model. Finally, use the benchmark’s analysis script to format results and average them with each model’s architecture. Acknowledgements This research was funded in part by the Estonian Research Council grant No. PRG2198 "MultiFlow – Multiscale Natural Flow Sensing for Coasts and Rivers” and the European Union via grant No. TEM-TA141. The authors express their gratitude to Maris Koppel, Jevgeni Pakin, Gert Toming and Helena Carmen Udu for their help and effort during data annotation. The data used in this study was selected from a larger dataset of over 380,000 videos, collected and manually annotated in the project "Fishway efficiency evaluation", funded by the German Ministry of Transport.

Fe-bearing clay minerals contain structural iron that canbe redox-active and can participate in electron transfer reactions withaqueous species. Although these redox properties have been studiedextens…

Fe-bearing clay minerals contain structural iron that canbe redox-active and can participate in electron transfer reactions withaqueous species. Although these redox properties have been studiedextensively in the past decade, questions remain about the respectiveroles of kinetic and thermodynamic constraints in establishing steadystateredox conditions. In this study, the reduction kinetics of aqueousCr(VI) to Cr(III) by Fe(II) contained in the structure of reducedferruginous clay samples (reference Nontronite NAu-1) wasmonitored with quick-XAS (X-ray absorption spectroscopy). Thesemeasurements revealed the occurrence of at least two reactionprocesses with contrasting fast and slow kinetic rates. According tomass and electron balance calculations, Fe(II) located at the edge ofthe clay mineral particles alone cannot account for the fast reactivity ofthe samples, pointing out the presence of electron transfer from theinner part of the clay mineral layer structure to the reactive sites. The Fe(II)/Fe(III) ratio in the clay structure quickly reached asteady state after each Cr(VI) addition to the solution. These steady-state conditions were consistent with either a completedepletion of the Cr(VI) reactant for the first spikes of Cr(VI) or a thermodynamic equilibrium between the redox couples, i.e.,between structural Fe(III)/Fe(II) and aqueous Cr(VI)/Cr(III), after the pool of fast-reacting Fe(II) was depleted. These resultshighlight the need to consider kinetic and thermodynamic controls of clay structural iron redox reactivity to predict the fate of redoxsensitivecontaminants in the environment.

This study examined the contribution of heads of schools' ethical modelling on promoting students' compliance with school rules and regulations in private secondary schools in Morogoro Municipality, T…

This study examined the contribution of heads of schools' ethical modelling on promoting students' compliance with school rules and regulations in private secondary schools in Morogoro Municipality, Tanzania. More specifically, the present study investigated the relationship between the ethical practice of the leadership and discipline and the responsible behaviour of the students through the following focus areas: hard work, punctuality, fairness, respectful discipline and promotion of moral values. The Convergent research design under the Mixed Methods Research Approach study was based on the Instructional Leadership Theory. Surveys and data analysis were conducted for the data collection process. The data collected by survey questionnaires were analysed quantitatively with descriptive statistics. Meanwhile, data obtained through semi-structured interviews and documentary sources were analysed with thematic analysis. Results indicated that ethical modelling has a positive contribution to students' willingness to abide by school rules and regulations. The study found that leaders and staff set an example which consistently emphasised hard work, punctuality, fairness, professionalism and respect, leading students to develop self-discipline, responsibility and positive attitudes towards school rules. Teaching in guidance and counselling programmes, moral education activities and fair disciplinary practices helped to further reinforce students' responsible behaviours and trust in school leaders. Ethical values were also found to be included in the schools' policies, guidance and counselling programmes, staff meetings, and disciplinary procedures, as evidenced by documentary records. The study concluded that modelling is an effective instructional leadership technique in fostering students' discipline, and it is recommended that school leaders consistently model ethical behaviour, enhance their moral education and counselling programmes, and treat everyone fairly and with respect when enforcing school rules.

The source code for our CCS 2026 paper: No One Cares About Privacy But You: Understanding Data Deletion Practices in the App-in-App Ecosystem.

The study examined the correlate between glycemic parameters (FPG, HBA1C, and INSULIN) and anthropometric parameters (BMI, WC, WHR, WHTR, BRI, AND VAI) among patients with type 2 diabetes mellitus. In…

The study examined the correlate between glycemic parameters (FPG, HBA1C, and INSULIN) and anthropometric parameters (BMI, WC, WHR, WHTR, BRI, AND VAI) among patients with type 2 diabetes mellitus. In carrying out this study comparative cross-sectional study was adopted. The study population consisted of one hundred (100) diagnosed type 2 diabetic patients and one hundred (100) apparently healthy, age- and sex-matched non-diabetic individuals who served as controls. All participants were aged between 30 and 70 years. Ten (10) mL of venous blood samples were collected from each participant after overnight fast using standard aseptic procedures and distributed into appropriate tubes for the determination of fasting plasma glucose (FPG), glycated hemoglobin (HbA1c), insulin, lipid profile, and oxidative stress markers. Serum and plasma were prepared by centrifugation at 3000 rpm for 10 minutes, while whole blood was used for selected analyses. The samples were analyzed using standard laboratory methods. Data obtained were analyzed using Statistical Package for Social Sciences (SPSS) version 25.0. The results revealed that fasting plasma glucose (FPG) had significant negative correlations with body mass index (BMI) (r = −0.216, p = 0.031) and waist-to-hip ratio (WHR) (r = −0.205, p = 0.041), while a significant positive correlation was observed between FPG and body roundness index (BRI) (r = 0.211, p = 0.035) among patients with type 2 diabetes mellitus. The study concluded that the observed relationships between glycaemic and anthropometric parameters indicate an association between glycaemic status and measures of body size and fat distribution among patients with type 2 diabetes mellitus. One of the recommendations made was that patients with type 2 diabetes mellitus should undergo regular assessment of FPG, HbA1c, and appropriate anthropometric parameters to facilitate effective monitoring of glycemic and metabolic status.

Brief overview over parliamentary system and participation in the European Semester (state of play March 2026).

Full Changelog: https://github.com/jp-textbook/jp-textbook.github.io/compare/v31.0...v31.1