识别模型loss - Githubissues

shawnh2 / BankCard-Recognizer

Identifying numbers from bankcard, based on Deep Learning with Keras

MIT License

80 stars 31 forks source link

Open fantasy955 opened 3 years ago

fantasy955 commented 3 years ago

请问您的识别模型当初训练了多久呀，我基于预训练的参数，loss最低降到2.6左右，之后上升。最终early stop。训练有什么技巧吗。谢谢。

shawnh2 commented 3 years ago

在crnn\cfg.py中尝试这组参数:

VALIDATION_RATIO = 0.2
BATCH_SIZE = 32 (或更大)

并从头（preprocess.py）开始