EVISAN: Expert-Annotated Sperm Images for Detection and Tracking

EVISAN contains 6,000 partially annotated microscope images of motile sperm from different donors and magnifications, creating a heterogeneous pool for deep learning. The dataset was manually annotated and checked by several biologists using semi-automatic software to ensure quality. It includes a standard evaluation framework for quantitatively assessing multi-target tracking and detection algorithms.

Use Cases

• [Training deep learning models for sperm detection] based on [6,000 annotated JPEG images]
• [Benchmarking multi-target tracking algorithms] based on [the provided standard evaluation framework]
• [Developing real-time computer-assisted sperm analysis (CASA) tools] based on [time-lapse microscopy sequences of motile cells]
• [Studying image heterogeneity in biological samples] based on [images from different donors under varying magnification]

Strengths

• Contains 6,000 images, described as an 'unparalleled heterogeneous dataset' for this specific application.
• Annotations were manually created and validated by several biologists, suggesting high label quality.
• Includes a standard evaluation framework, facilitating reproducible benchmarking of algorithms.

Limitations

• Specific column names and the exact data structure within the XML files are not detailed.
• The dataset size is only described by image count; the number of annotated objects or sequences is unknown.
• Metadata such as the last update date and precise file sizes are missing from all sources.

Provenance

Source

Tsinghua University, authored by Yimo Yan.

Collection Method

Semen samples from different donors were imaged using a microscope under varying magnification. Sperm images were manually annotated and checked by biologists using semi-automatic software.