請問這類的驗證碼的樣本數和配置 #106
Description
請教一下，目前驗證碼的特徵有： 斜體字 / 前景干擾 / 背景干擾 / 多種字體 / 數字
目前配置
Label Num: 5
CNN Layer: CNN3
Recurrent Layer: GRU
UnitsNum: 64
Loss Function： CTC
Optimizer: RAdam
Learning Rate: 0.001
Category: Numeric
Epoch已經1864但是Accuracy仍然是0，訓練樣本數是700張，用來驗證有300張
請問可以調整配置，讓Accuracy上來
Log
INFO:tensorflow:0 5 3 [6, 9, 3, 6, 9] [5, 6, 1, 11, 11, 11] --> ['5', '8', '2', '5', '8'] ['4', '5', '0', '-', '-', '-']
INFO:tensorflow:1 5 2 [7, 10, 8, 1, 3] [8, 1, 11, 11, 11, 11] --> ['6', '9', '7', '0', '2'] ['7', '0', '-', '-', '-', '-']
INFO:tensorflow:2 5 4 [3, 7, 10, 1, 9] [2, 9, 1, 7, 11, 11] --> ['2', '6', '9', '0', '8'] ['1', '8', '0', '6', '-', '-']
INFO:tensorflow:3 5 2 [7, 8, 4, 1, 6] [10, 6, 11, 11, 11, 11] --> ['6', '7', '3', '0', '5'] ['9', '5', '-', '-', '-', '-']
INFO:tensorflow:4 5 5 [5, 8, 6, 9, 5] [9, 5, 11, 2, 6, 10] --> ['4', '7', '5', '8', '4'] ['8', '4', '-', '1', '5', '9']
ERROR:tensorflow:[{"origin": "58258", "decode": "450---"}, {"origin": "69702", "decode": "70----"}, {"origin": "26908", "decode": "1806--"}, {"origin": "67305", "decode": "95----"}, {"origin": "47584", "decode": "84-159"}]
INFO:tensorflow:Epoch: 1864, Step: 20500, Accuracy = 0.0000, Cost = 0.08694, Time = 1.371 sec/batch, LearningRate: 0.00048020004760473967