Closed Bercraft closed 6 months ago
You can generate captions with other tools out there and my tools will be able to deal with .caption text files (although it's limited at this moment).
I explained in this post why models like BLIP currently aren't supported: https://github.com/Particle1904/DatasetHelpers/issues/14#issuecomment-1951163757
I've tried before but I'm simply lacking the necessary knowledge to implement inference for caption models that use tokenization. Furthest I got was converting a model like BLIP to .onnx format. I could be lazy and use interop with Python... and C# is great at it, but I personally I'm not using the tools that much lately and it's mostly not being developed anymore, unless I need a new feature; which happens very rarely nowdays.
Thank you i understand.
Could be possible to add a something like BLIP?? thank you