Hello all !
I am looking for a github code of image classification which take large images (1500x2000 pixels) images at input of neural network (all codes I have found take maximum 224x224 pixels images) and which take account of very imbalanced classes. I must can input my own dataset for training.
Hello all !
Welcome to the TensorFlow Forum!
You can use the same image classification model for the training with the large input_shape and with imbalanced classes. You may need to enable GPU for the computations because of the large image shape takes more memory space and it may take plenty of time for the model training and also because of imbalanced classes dataser, the trained model may not predict better.
Hence we recommend to use the balanced classes dataset and should not use larger input_shape so that model can be trained more efficiently and provides better outputs.
You can preprocess the dataset to resize, rescale or crop the images before feeding into the model as well as can use data augmentation technique to balance the dataset. Please have a look at this doc for the reference. Thank you.