Accuracy limitations for model deployment

I am training a logo detection model using resnet50 and the logo 2k+ dataset. My model has achieved an accuracy of about 0.65 which is close to the results recorded in the paper for the dataset ( Logo-2K+: A Large-Scale Logo Dataset for Scalable Logo Classification) for resnet50. I would love to know if the current accuracy I have is too low for deployment in a logo detection web service. I plan to add features to enable the model learn from the new data it collects from users who use the service. Please any ideas on how to improve the accuracy will also be appreciated! Thank you.