configs:
- config_name: default
data_files:
- split: fold_0
path:
- data/1.avi
- data/2.avi
- data/3.avi
- data/4.avi
- data/6.avi
- data/7.avi
- data/8.avi
- data/9.avi
- data/11.avi
- data/13.avi
- data/14.avi
- data/16.avi
- data/17.avi
- data/18.avi
- data/20.avi
- data/24.avi
- data/27.avi
- data/metadata.csv
- split: fold_1
path:
- data/5.avi
- data/10.avi
- data/12.avi
- data/15.avi
- data/19.avi
- data/21.avi
- data/22.avi
- data/25.avi
- data/26.avi
- data/31.avi
- data/32.avi
- data/33.avi
- data/34.avi
- data/36.avi
- data/37.avi
- data/39.avi
- data/41.avi
- data/metadata.csv
- split: fold_2
path:
- data/23.avi
- data/28.avi
- data/29.avi
- data/30.avi
- data/35.avi
- data/38.avi
- data/40.avi
- data/42.avi
- data/43.avi
- data/44.avi
- data/45.avi
- data/46.avi
- data/50.avi
- data/53.avi
- data/58.avi
- data/64.avi
- data/65.avi
- data/metadata.csv
- split: fold_3
path:
- data/47.avi
- data/48.avi
- data/49.avi
- data/51.avi
- data/52.avi
- data/54.avi
- data/55.avi
- data/56.avi
- data/57.avi
- data/60.avi
- data/61.avi
- data/62.avi
- data/63.avi
- data/67.avi
- data/70.avi
- data/72.avi
- data/73.avi
- data/metadata.csv
- split: fold_4
path:
- data/59.avi
- data/66.avi
- data/68.avi
- data/69.avi
- data/71.avi
- data/74.avi
- data/75.avi
- data/76.avi
- data/77.avi
- data/78.avi
- data/79.avi
- data/80.avi
- data/81.avi
- data/82.avi
- data/83.avi
- data/84.avi
- data/85.avi
- data/metadata.csv
task_categories:
- video-classification
- image-feature-extraction
language:
- en
tags:
- sperm
- sperm-video
- morphology
- mortility
pretty_name: VISEM
size_categories:
- n<1K
Dataset Card for VISEM Dataset
Dataset Details
Dataset Description
The VISEM dataset is a multimodal video dataset designed for the analysis of human spermatozoa. It is one of the few open datasets that combine multiple data sources, including videos, biological analysis data, and participant-related information. The dataset consists of anonymized data from 85 different participants, with a focus on improving research in human reproduction, particularly male fertility and semen quality analysis.
- Curated by: Simula Research Laboratory and Oslo Metropolitan University
- Funded by [optional]: Norwegian Research Council (ReproAI project, project number 259293/o70)
- Shared by [optional]: Simula Research Laboratory
- Language(s) (NLP): Not applicable
- License: The dataset can be used for research purposes. Commercial use requires approval from the authors.
Dataset Sources [optional]
- Repository: VISEM Dataset Repository
- Paper [optional]: VISEM: A Multimodal Video Dataset of Human Spermatozoa
- Demo [How to load this dataset and extract a frame]: Jupyter-notebook
Uses
The use of the VISEM dataset is restricted to research and educational purposes. Commercial use requires prior written permission. Any publications using or referencing the dataset must cite the dataset paper: https://dl.acm.org/doi/10.1145/3304109.3325814.
Direct Use
The dataset is useful for research in male fertility assessment, machine learning-based semen analysis, and medical image/video processing. Some key applications include:
- Automatic sperm tracking in real-time
- Predicting sperm motility and morphology from video data
- Correlating semen quality with other biological parameters
- Enhancing computer-aided sperm analysis (CASA) with deep learning techniques
A Sample Notebook showing how to load and extract a frame from this dataset is here.
Out-of-Scope Use
- Commercial use without prior approval
- Any use that violates ethical guidelines for medical and biological research
Dataset Structure
The dataset consists of over 35GB of video data and six CSV files containing:
- Semen analysis data – Standard semen parameters (e.g., sperm motility, sperm count)
- Fatty acids in spermatozoa – Levels of various fatty acids in sperm cells
- Fatty acids in serum – Levels of phospholipids from blood samples
- Sex hormones – Hormone measurements from participants' blood samples
- Participant-related data – Age, BMI, abstinence time, and other demographic information
- Video metadata – Mapping between videos and participant IDs
Each participant has one associated video file, with a resolution of 640×480 pixels and a frame rate of 50 fps.
Dataset Creation
Some scripts used to split data into folds in a stratified way: Sperm-Net
Curation Rationale
The dataset was created to improve the understanding of male fertility factors and to encourage interdisciplinary research in multimedia, machine learning, and medical informatics.
Source Data
Data Collection and Processing
- Participants were recruited between 2008 and 2013.
- Standard semen analysis followed WHO guidelines (2010 edition).
- Videos were recorded using a UEye UI-2210C camera mounted on an Olympus CX31 microscope.
- Fatty acid composition was measured via gas chromatography.
- Sex hormones were analyzed using standardized blood tests.
Who are the source data producers?
The dataset was collected by researchers from Simula Research Laboratory and Oslo Metropolitan University, in collaboration with medical professionals specializing in reproductive health.
Annotations [optional]
Annotation Process
- Standard WHO semen analysis was performed manually by laboratory personnel.
- Motility categories (progressive, non-progressive, immotile) were assigned based on manual assessments.
- Morphological classifications were determined according to established medical guidelines.
Who are the annotators?
Qualified medical and laboratory professionals specializing in semen analysis.
Personal and Sensitive Information
All data has been anonymized. The dataset does not contain personally identifiable information but does include demographic and biological data.
Bias, Risks, and Limitations
- The dataset includes only 85 participants, which may limit generalizability.
- Participants were recruited primarily from Norway, which may introduce geographic bias.
- Some dataset attributes (e.g., sperm motility) may vary significantly between individuals, requiring careful interpretation.
Recommendations
Users should be aware of dataset biases and limitations. Future expansions with larger and more diverse populations are recommended to improve generalizability.
Citation [optional]
BibTeX:
@inproceedings{haugen2019visem,
author = {Haugen, Trine B. and Hicks, Steven A. and Andersen, Jorunn M. and Witczak, Oliwia and Hammer, Hugo L. and Borgli, Rune and Halvorsen, Pål and Riegler, Michael},
title = {VISEM: A Multimodal Video Dataset of Human Spermatozoa},
booktitle = {Proceedings of the 10th ACM Multimedia Systems Conference},
year = {2019},
pages = {6},
doi = {10.1145/3304109.3325814}
}
APA:
Haugen, T. B., Hicks, S. A., Andersen, J. M., Witczak, O., Hammer, H. L., Borgli, R., Halvorsen, P., & Riegler, M. (2019). VISEM: A Multimodal Video Dataset of Human Spermatozoa. Proceedings of the 10th ACM Multimedia Systems Conference. https://doi.org/10.1145/3304109.3325814
Glossary [optional]
- CASA: Computer-aided sperm analysis
- Motility: The ability of sperm to move effectively
- Morphology: The shape and structure of spermatozoa
- Teratozoospermia Index: A measure of abnormal sperm morphology
More Information [optional]
For further inquiries, visit the dataset repository or read the original paper.
Dataset Card Authors [optional]
- Compiled by ChatGPT based on official sources
Dataset Card Contact
For inquiries regarding the dataset, contact Simula Research Laboratory at Simula Contact.