Skip to content

Releases: VectorInstitute/SegMate

SegMate v0.3.1

26 Oct 16:47
Compare
Choose a tag to compare

Bug fix:

  • Updated default value for point prompts updated in encode_input function
  • Fix the prompt dimension error in finetune function
  • Updated BIS dataset box prompt shape

SegMate v0.3.0

26 Oct 16:41
Compare
Choose a tag to compare

SAM prompt tuning:

  • Added example showcasing how to perform prompt tuning on satellite images using SegMate
  • Added functions in SAM class and updated util functions to support prompt tuning
  • Updated fine-tuning code to be more generalized and support different prompt types

SegMate v0.2.0

26 Oct 16:32
Compare
Choose a tag to compare

Restructured functions and classes:

  • The SegMate class is now split into Segmented parent class and SAM child class, and object detector is no longer a class attribute
  • Moved all non-model related functions to utils
  • Removed some unused functions
  • Updated some utils functions for more flexibility
  • Added deforestation demo, and example training a custom decoder to auto-segment specific objects
  • Added support for mask prompt

SegMate v0.1.0

26 Oct 15:18
Compare
Choose a tag to compare

First release of the SegMate package, main features include:

  • Easy inference of SAM model with point prompts, box prompts, or auto-segmentation
  • Added text prompt segmentation capability for SAM with the support of GroundingDINO
  • Fine-tuning function from the provided BIS dataset