Fine-tuning about crnn-pytorch

[Holmeyoung]

Hi, you should change train.py to

import torch.nn as nn
from models.crnn import BidirectionalLSTM
def net_init():
    nclass_pre = 11 # the nclass of your pre-trained model:  = len(params.alphabet--pre version) + 1
    nclass = len(params.alphabet) + 1
    crnn = net.CRNN(params.imgH, params.nc, nclass_pre, params.nh)
    crnn.apply(weights_init)
    if params.pretrained != '':
        print('loading pretrained model from %s' % params.pretrained)
        if params.multi_gpu:
            crnn = torch.nn.DataParallel(crnn)
        crnn.load_state_dict(torch.load(params.pretrained))
    
    crnn.rnn = nn.Sequential(
            BidirectionalLSTM(512, params.nh, params.nh),
            BidirectionalLSTM(params.nh, params.nh, nclass))
    return crnn

[Holmeyoung]

Hi,

1. load the pre-trained model use the same net as the pre-trained model

crnn = net.CRNN(params.imgH, params.nc, nclass, params.nh)

here, the nclass should't equal to len(params.alphabet) + 1, it should be the classed number of pre-trained model.

2. change the last layer to yourself.

crnn.rnn = nn.Sequential(
            BidirectionalLSTM(512, nh, nh),
            BidirectionalLSTM(nh, nh, nclass))

nclass = len(params.alphabet) + 1

[niddal-imam]

Thanks for the quick response.
Now I should first load the pre-trained model by changing the params.py
pretrained = 'path/to/my/pre-trained'
But I did not understand the second point. What should I change?

Thanks

[Holmeyoung]

After load the model, change the rnn layer.

[niddal-imam]

Thank you Holmeyoung. Should I change crnn.py:
self.cnn = cnn self.rnn = nn.Sequential( BidirectionalLSTM(512, nh, nh), BidirectionalLSTM(nh, nh, nclass))
to
crnn.rnn = nn.Sequential( BidirectionalLSTM(512, nh, nh), BidirectionalLSTM(nh, nh, nclass))?

[niddal-imam]

Thank you very much.