search

visual-layer / fastdup

fastdup is a powerful free tool designed to rapidly extract valuable insights from your image & video datasets. Assisting you to increase your dataset images & labels quality and reduce your data operations costs at an unparalleled scale.
Other
1.52k stars 74 forks source link

Update v1 api to support image search #276

Closed dnth closed 9 months ago

dnth commented 10 months ago

This PR adds image search support to the V1 API.

Closes https://github.com/visual-layer/fastdup/issues/269

dbickson commented 9 months ago

@dnth the code will not support fully model_paths that are allowed resent50, efficientnet, clip14, clip336 etc etc I prefer not to include onnxruntime here, if the user has no standard model all new resize param

dbickson commented 9 months ago

@dnth I merged this manually, please note the function documentation was missing i copied it from fastdup.init, in addition the user should resize the image on his own. I also added fd.vector_search() which is support for low level vector search (instead of image search)