Is there any way to train with custom data?
Hi I'd like to ask is there any plan to make an description or example code for custom data train. Or Can I just ask is it possible to train with my own data?
Hi Kyle, it's definitely possible to train with your own data. Unfortunately, we currently don't expose our training pipeline
Thank you for your reply
Then, is there a way to train my own data without exposing the training pipeline?
Take a look at IS-Net - that's the baseline of the training architecture
If fine-tuning still interests you, explore my GitHub repository. You'll find a notebook there that demonstrates how to train a model on one picture. It's just a raw exaple, but I hope It will be useful.
I recommend first trying to apply the original model to an input picture, and then initiating the fine-tuning process. The improvement targets are beyond just facial enhancements.
Please note the line model_weights_path = f"finetuned_model.pth" in example_inference.py. You should replace this with the path to the specific model you intend to use.
Regrettably, GitHub restrictions prevented me from uploading the model directly. However, you can download it from the original author's repository
https://github.com/WpythonW/RmBgModel.git