Comments (20)
数据和现在训练使用的模型参数大概是什么?
from sightseq.
densenet121,其他参数基本默认,loss可以下去,但valid data不好,数据的话是一条条中文大写金额
from sightseq.
先把模型换成 densenet_cifar
试一下呢, densenet_cifar
把图片放缩 1/4,densenet121
放缩 1/8。1/8 有时候放缩得太严重了。
from sightseq.
好的,我想再问下,空格是无需打上label吧?
from sightseq.
不需要
from sightseq.
BTW, --height
, --width
也要和训练使用的图片保持比例
from sightseq.
thx, 这个之前注意到了,现在在trainning过程中,validation acc就不高 ,真实数据train 2000多张,test200张,densenet121 最好的一次是acc 23%。场景下的训练,不知道大概需要多少量的训练数据
from sightseq.
先换成 densenet_cifar
跑一下。如果 validate 的数据和 train 的差距不大,2000 张感觉也够用训练,transformer 的 mean
, std
这个需要改一下
from sightseq.
我之前在你代码基础上加入了多gpu模式可以了,transformer 的 mean, std 是需要做什么样的修改?
from sightseq.
使用 train 的数据的 mean
和 std
,小数据集这个还挺关键的. 多 gpu 模式欢迎提 pull request
from sightseq.
不好意思,可能初学不是特别明白,transform 的 mean 和 std 统计的是什么?
from sightseq.
我描述可能不太清楚。是使用的图片的均值和方差, 在 main.py 的这 两行. 图片送进网络模型的时候先经过 transformer
这一步。
可以参考 script 计算均值和方差.
from sightseq.
代码位置我看到了,参考代码也看了下,可以简单理解为 图片像素的均值和方差吗? 然后现在main中的的mean 和std 是加载 模型自带的值吗
from sightseq.
是的, crnn.py 里面我写了我自己一个数据的 mean
和 std
from sightseq.
嗯,在前面做下归一化,图像预处理部分。如果target 的像素分布和source 相似的话,是有用额
from sightseq.
您这边能否share一个比较好的 图片数据增广的code, 我之前是用了text_render文本生成图片,我还想找适合crnn训练用的, 能给定一个图片,然后通过各种变化生成新图的,传统的图像方法的code,非DCGAN
from sightseq.
我用过这个 https://github.com/mdbloice/Augmentor
这个没用过 https://github.com/aleju/imgaug 但看起来还不错
from sightseq.
Augmentor这个不错,就是没有色彩、对比度这种的变换
第二个用起来比较重
from sightseq.
@peiji1981 不知道你那边结果怎么样?
CTCLoss 官方现在建议更新最新的 pytorch,CTCLoss 实现有 bug, 讨论在 pytorch/pytorch#21392
from sightseq.
造了点数据有一定提升,嗯 我更新下CTCLOSS
from sightseq.
Related Issues (20)
- Help Needed HOT 7
- Questions about dataset object HOT 4
- RuntimeError: CUDA error: an illegal memory access was encountered HOT 7
- loss become inf , then Nan HOT 18
- Input size HOT 1
- dimensions in forward pass HOT 4
- 中文识别率不高是不是因为感受野的原因? HOT 16
- 有关loss变为nan的情况,我看了之前的解答,但还是想问问 HOT 2
- Getting accuracy as 0.00 HOT 1
- 关于加载预训练模型的问题 HOT 1
- How is the picture processed in sequence_generate? HOT 1
- 能提供新的依赖版本么? HOT 5
- TypeError: 'DigitsBatchTrain' object is not iterable HOT 15
- The vanilla cnn downsampling architecture cannot recover spatial information of a image HOT 1
- annotation file format for English data HOT 2
- Not found recurrent layer in model files HOT 1
- Must the training data be of equal length? HOT 1
- 训练结果在其他图片上的结果很差? HOT 3
- 同一批测试数据,test-only 的accuracy和 训练时的validate accuracy 差很多? HOT 4
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from sightseq.