frotms comments

Results 122 comments of


                                            frotms

The real Params in my experiment

> There are 33.6M when I use the 'saved_model' pattern to save the mobilenetv3_large_1.0 models. Is there anything wrong that I can deal with? It's about 31M when I just...

something wrong about training

try cuda()

Dataset

what's your email? @bopeng87 @fanhaidi

Training loss

You lost parameters of batch normalization layer. @weishengchong ``` with tf.control_dependencies(tf.get_collection(tf.GraphKeys.UPDATE_OPS)): train_ops = optimizer.minimize(loss) ```

网盘密码

写了呢，p018

后续更新

暂时没计划

PaddleOCRV4小语种模型

您可以参照V4的转换更新小语种模型

繁体字V3模型效果很差

@1556900941lizerui 1、本身v3模型转换有精度损失 2、老版的预处理和后处理代码可能和现在版本的paddle没对齐

vi-layoutxlm权重转pytorch

目前不支持

AttributeError: module 'paddle.fluid' has no attribute 'load_dygraph'

可能是新版paddle去掉这个接口了