crnn.pytorch
crnn.pytorch copied to clipboard
meijieru
Convolutional recurrent network in pytorch
我用create_dataset.py处理了自己的训练数据，然后放入CRNN中训练，出现以下错误： /crnn_net/dataset.py:95: UserWarning: torch.range is deprecated in favor of torch.arange and will be removed in 0.5. Note that arange generates values in [start; end), not [start; end]. batch_index = random_start...
While training the model at every epoch the loss is always a nan or inf value, I tried to debug but could not sort out. If anyone face this kind...
I noticed that this project and several more use warp_ctc_pytorch.ctcloss instead of torch.nn.ctcloss, however, installing warp_ctc_pytorch library did not work with me, so may I know what is the different...
在做CRNN数字辨识时，用mnist的资料集来随机生成不等长数字。总共生成过6w和10w笔资料。损失值到0.00005停止，都发生过拟合现象，验证集准确率很高，但是测试准确率很低，是因为只用mnist来当训练集过于单一吗？ In the process of crnn number identification, the data set of MNIST is used to generate unequal numbers randomly. A total of 6W and 10W data were generated. When...
Hello @meijieru and collaborators, I'm trying to use this code but I'm having some troubles with it, which led me to try and step through the training code and see...
请问有人用这个训练过论文里用的数据集ICDAR2003，ICDAR2013，IIIT5k，SVT嘛？准确率是多少？ 我用这个训练IC13准确率只有百分之六十多，但是论文里是百分之八十多啊。。。
Ubuntu 17.04 no gpu python 3.6 cs316@xxxx:~/lijian/crnn.pytorch-master$ python train.py -- trainRoot ./data/train --valRoot ./data/val --random_sample --workers 1 Namespace(adadelta=False, adam=False, alphabet='0123456789abcdefghijklmnopqrstuv wxyz/', batchSize=64, beta1=0.5, cuda=False, displayInterval=500, expr_dir='expr ', imgH=32, imgW=100, keep_ratio=False,...
my accuracy is always 0, i have saw the feature which crnn net outputs, and after max it ,i find that almost all indexs of labels are 0, which means...
