FishIndivID: Zebrafish individual identification and tracking online platform

Overview

This platform is a specialized tool for individual zebrafish identification based on images, supporting users in uploading images from two time points to achieve cross-developmental-stage individual matching. Through dedicated models targeting different body parts and developmental stages, it enables long-term identity tracking. The platform currently offers four pre-trained recognition models. The modeling data comes from two zebrafish developmental stages (Stage 1: 30-40 dpf, Stage 2: 50-60 dpf) and two body parts (lateral view, dorsal head view). Additionally, the platform provides reference accuracy thresholds for each model across different time intervals. Users can select the appropriate model based on the developmental stage of the zebrafish to be identified, the photographed body part, and the time interval between the two uploaded images.

FishIndivID is a platform enabling marker-free, non-invasive long-term tracking of individual zebrafish based on biological features. It adapts to zebrafish growth changes through staged models that address morphological variations during individual development. Users can directly utilize pre-trained models for identification without requiring hardware investment. To view accuracy curves for the models across different time spans or to submit batch identification requests, please visit the Help page or contact us. The platform is continuously expanding its species recognition capabilities and will support more zebrafish developmental stage models in the future.

If you have any questions about FishIndivID, contact us at xqxia@ihb.ac.cn.

• The figure above shows the overall process of zebrafish individual identification.
• ESCAlignNet performs target detection on zebrafish and performs segmentation and alignment standardization. And IDNet performs feature extraction on zebrafish images.
• The Image Database stores and registers standardized images for individual identification management. And the result determines the individual identity based on the threshold and outputs it.

Reference of Time Span

• The figures illustrate the effective duration during which four models can maintain recognition accuracy above 95% from Time 1 to Time 2.
• Users may set an expected recognition accuracy threshold in the interface. Upon submission, the system will display the effective duration between any two time points (within 31-122 dpf) for all 4 models when performing identification based on either the lateral body region or dorsal head region while sustaining the specified accuracy threshold.