Low Accuracy

[DennisFaucher]

Thank you for this excellent example. Great idea. I have found the accuracy of the model quite low for me (~20%) and incorrect letters are often suggested. I decided to create a new model with your huge dataset in the NVIDIA DIGITS program using AlexNet. Accuracy jumped to 100% in just the first few epochs of training. I will try and adjust your code to use the Caffe model that DIGITS creates.

[csyjin6]

how do you do this?

[DennisFaucher]

how do you do this?

The tutorial on DIGITS is very good. Take a look at the Digits section of "Two Days to a Demo" on this page: https://github.com/dusty-nv/jetson-inference/blob/master/README.md#two-days-to-a-demo-digits

[vkravchik]

how do you do this?

The tutorial on DIGITS is very good. Take a look at the Digits section of "Two Days to a Demo" on this page: https://github.com/dusty-nv/jetson-inference/blob/master/README.md#two-days-to-a-demo-digits

Hey, can u help with that? I need more accuracy for my graduated work

[ChamiJ]

Thank you for this excellent example. Great idea. I have found the accuracy of the model quite low for me (~20%) and incorrect letters are often suggested. I decided to create a new model with your huge dataset in the NVIDIA DIGITS program using AlexNet. Accuracy jumped to 100% in just the first few epochs of training. I will try and adjust your code to use the Caffe model that DIGITS creates.

Hi could you please guide me how to get the accuracy graphs for the loicmarie's model?please provide me the code.

[DennisFaucher]

@ChamiJ NVIDIA DIGITS creates those graphs https://developer.nvidia.com/digits