Data format
What to send SymNexusVision
The inputs the platform expects, and the labels it learns from.
Input
Input format
Images in common formats (PNG, JPG, TIFF), one object scene per image.
| file | width | height |
|---|---|---|
| site_cam_0412.jpg | 1920 | 1080 |
| plate_A03.tif | 2048 | 2048 |
file,width,height
site_cam_0412.jpg,1920,1080
plate_A03.tif,2048,2048For training
Label format
For training: outlines (polygons) or boxes around each object with its type, in the standard COCO JSON format; ImageJ/Fiji ROI files can be imported directly.
| image | object type | outline (x,y points) |
|---|---|---|
| plate_A03.tif | worm | [[412,88],[430,91],…] |
| site_cam_0412.jpg | hard hat | [[902,211],[955,209],…] |
Tips for good results
- Include the lighting, angles and backgrounds you'll see in real use.
- Outline every object of interest in each labelled image.
- Start with 200-500 labelled images per object type.