Does the project type really matter or is it just for information? Does selecting object detection or classification matter when creating a new project/does it change the workflow/output of the trained model?
Yes, it does matter. This is cemented when you create the project and cannot be changed later. If this is a work flow issue you can create an Object Detection project first and then create a 2nd one and use classification
have a good tima @anon59033456 , I have chosen instance segmentation instead of semantic segmentation for the typing section of a project that I want to use U-Net architecture for. Is there a way to change the typing project? I have annotated the entire dataset.
@Navid_Mashoofi That option is available, but we do now have an easier process for conversion between the two if you’ve mistakenly labelled in the incorrect project type.
Semantic Segmentation to Instance Segmentation
- Generate a raw-image version (no preprocessing or augmentation other than Auto-Orient, to ensure all labels remain aligned with the image’s orientation during labelling)
-
Export in COCO Segmentation format
-
Create a new Instance Segmentation project
-
Unzip the COCO Segmentation export from your Semantic Segmentation project, and upload to the new project
Instance Segmentation to Semantic Segmentation
-
Follow steps (1) and (2) above
-
Create a new Semantic Segmentation project
-
Follow step (4) from above
Dataset export:
Annotation formats:
Dataset upload options:
- available via UI
- CLI: Command Line Interface - Roboflow
- API: Upload API - Roboflow | hosted images
- Python Package: Upload API - Roboflow
More on project types:
- Difference Between Semantic and Instance Segmentation **
- What is Instance Segmentation? A Guide.
- What is Object Detection? (YouTube video)
- What is Image Classification? (YouTube video)
- YouTube video on computer vision model types (Roboflow YouTube): Computer Vision Model Types - YouTube
- Introduction to Computer Vision **
Your response was incredibly informative and provided me with the clarity I needed to move forward with my project. Thank you again for your generosity and willingness to share your expertise. @Mohamed