Zebrafish
Description
The annotated dataset is completed using images in the mixed mode, which contains 14,042 zebrafish samples and 3512 negative samples.
Files
Steps to reproduce
The process of image acquisition is as follows: (1) The event camera is aimed directly at a breeding tank containing 20 zebrafish to collect the event stream data. (2) When the collection is finished, the event stream data is converted into videos in four modes: binary, count, gray, and accumulate. Then, videos of these four modes are fused to generate a mixed mode video, resulting in five videos with the same content. (3) For these five videos, image frames are extracted every 5 seconds. Since the zebrafish swim randomly, the fixed time interval will not affect the samples’ representativeness. In dataset images acquirement, the whole capturing process lasts for 60 minutes. For each video mode, 720 images are extracted. The LabelMe tool is used for manual labeling. Each target is marked with a bounding box. Zebrafish are labelled as class “zebrafish” and Reflections of zebrafish are labelled as class ‘negative sample’. The annotated dataset is completed using images in the mixed mode, which contains 14,042 zebrafish samples and 3512 negative samples. Since the five modes of videos are aligned in time axis, the position of each bounding box is almost the same for the five modes. Local adjustments will be made according to the actual situation.
Institutions
- Shanghai Dianji University