Spaces:
Running
Attempting to replicate results from huggingface on Google Colab
Hi Guys,
I am trying to ascertain what are the exact parameters used in huggingface demo versus your Google Colab implementation.
On the HuggingFace your model was able to detect all the text on an image using the db_resnet50 model. However, when I tried to replicate the same parameters in Colab the text detection does not find all text on images. I also applied the binarization of 30% on the same image.
Is there something else that I am missing or are using a updated version of db_resnet50 ?
Colab list the db_resnet50 model as https://doctr-static.mindee.com/models?id=v0.7.0/db_resnet50-79bd7d70.pt&src=0
Hi,
It was happening to me but with different models, demo had better precision with the same parameters, after I changed from tensorflow to torch results are the same as the demo
Yeah, I had pytorch set as the model but I found the previous model was better.