Project Setup
Configure dataset name, output location, and annotation format.
Project Details
Required — a project name is needed to save and organize your dataset.
Output Directory
A folder name is saved under the server's DatasetStudio datasets/ workspace. An absolute path is used as-is (in Docker it must be under the mounted volume to appear on the host). Leave blank to auto-generate a timestamped folder.
Annotation Format
Each image has a paired .txt file with normalized [class x y w h] per line. Compatible with Ultralytics YOLOv8+.
SAM3 Integration
Segment Anything Model 3 — used for background cleaning and object extraction
Detect & inpaint foreground objects from rasters to produce clean backgrounds
Click a pixel; SAM3 returns the contour mask of the object at that location
Draw a rectangle prompt; SAM3 tightly segments the enclosed object
Click one object; SAM3 finds all similar instances in the 1024px tile
Describe an object class in natural language; SAM3 segments by text
Interactive mask correction on generated tiles