Hello,
I need some advice and help from everyone who are welcoming to share their knowledge.
Actually I have only around 1000 image and I want to use them to generate this mode please check the link[the model I need to adapt to train 1000 image https://github.com/biswassanket/synth_doc_generation
Is it possible and how I can adapted the code to my small amount of data? Thank you in advance. .
Hello,
I need some advice and help from everyone who are welcoming to share their knowledge.
Actually I have only around 1000 image and I want to use them to generate this mode please check the link[the model I need to adapt to train 1000 image https://github.com/biswassanket/synth_doc_generation
Is it possible and how I can adapted the code to my small amount of data? Thank you in advance. .
Have you considered fine-tuning a pre-trained model?
hell fpozzi,
Thank you very much for your suggestion,
Please I have some questions to help me out as I am a beginner :
so for my case I have to use 1. trained model on PubLayNet
my data has the same classes as publaynet only with different language (not in english) it will work right ?
second any suggestion on how I fine-tuning it to adapt my data ?
Thank you.
hell fpozzi,
Thank you very much for your suggestion,
Please I have some questions to help me out as I am a beginner :
so for my case I have to use 1. trained model on PubLayNet
my data has the same classes as publaynet only with different language (not in english) it will work right ?
I think the best thing to do is to try. Another thing you can do is to search for similar works.
second any suggestion on how I fine-tuning it to adapt my data ?
Since your dataset is very small, you can freeze the first layers of your network and make the remaining layers trainable.