hiarindam / document-image-classification-tl-sg Goto Github PK

I'd like to use this algorithm to train some other ticket classification task. Could you offer the train code?

Other than pretrained weights, are you planning to also provide the reference source code (preprocessing, meta-classifier, etc...) used for the article?

Thanks

I computed mean and std for rvlcdip to be 0.9919 and 0.1853 so these are the following transforms I'm using for the train set:
dataset = RvlCdipDataset('labels/train.txt', 'images/', transform=transforms.Compose([transforms.Resize((224,224)), transforms.Normalize(0.9199, 0.1853), transforms.Lambda(lambda x: x.repeat(3,1,1))]))

Do I use the same transforms for the test set?

Dear author,

Please allow me to clarify some details regarding the implementation. To run inference on an image using your proposed "MLNN based stacking of holistic & region-based models with inter and intra-domain weights transfer" method, I would first run each of the 5 base models (with your given weights) on the image and get 5 different "base class softmax predictions". I would then concatenate those 5 base class softmax predictions... but since each softmax prediction has 16 scores wouldn't their concatenation have a total of 5*16 scores. Afterwards, I am supposed to feed those scores to an MLNN to get the final prediction. Is that right? Do you also have the weights for the MLNN? Or do I have to train it on the validation set myself?

Please help me as I would like to be able to reproduce your results (92.21% accuracy) for my research project.

Yours sincerely,
Gordon

It was never specified in the paper.

I see you post the link to pre-trained weights but no tutorial how to use them.

Tried with keras:

from keras.models import load_model
load_model('vgg16_weights_th_dim_ordering_th_kernels_Holistic_91.11.h5')

but failed ValueError: No model found in config file.

How do I load these h5 weights? Can you provide the model source so I can use model.load_weights(...)