Comments (8)
@hezw2016 1.About the init training learning rate it my mistake to write a wrong value in the readme file. The model was trained with the initialized lr 0.002. About the D_loss value problem you may put more details here to discuss it maybe some console print info for example:)
2. When I train my model the data augmentation was not used at all. The reason why I not use it is simply that I do not want to write those code==!. The cropped patches can be used as training images here and the method is quite efficient when I implement another derain task in the paper "Learning Dual Convolutional Neural Networks for Low-Level Vision"(http://faculty.ucmerced.edu/mhyang/papers/cvpr2018_dual_cnn.pdf). If you want to use the cropped patches as training samples you'd better the patches you cropped contain valid rain drop there:)
from attentive-gan-derainnet.
@MaybeShewill-CV
Thx for your reply.
I will re-train the model to see whether the problem still exists or not. If so, I will provide some details. : )
from attentive-gan-derainnet.
@hezw2016 If you re-train a model with the cropped patches you're welcome to share the result here. It may also help me and other people:)
from attentive-gan-derainnet.
@MaybeShewill-CV OK, no problem, it may take several days on my 1070 GPU. HaHa
from attentive-gan-derainnet.
@MaybeShewill-CV Hi, the D_loss problem happens again.
At the begining, D_loss is around 15.94, then the D_loss suddenly becomes very large.
here is some details:
++++++++++++++++++++++++
I1112 12:18:20.574566 21056 train_model.py:209] =====>Epoch: 6676 D_loss: 225723.28125 G_loss: 1.69909 Ssim: 0.76703 Cost_time: 1.06179s
I1112 12:18:21.646596 21056 train_model.py:209] =====>Epoch: 6677 D_loss: 120077.87500 G_loss: 1.64597 Ssim: 0.79087 Cost_time: 1.07203s
I1112 12:18:22.716292 21056 train_model.py:209] =====>Epoch: 6678 D_loss: 82806.89844 G_loss: 2.30736 Ssim: 0.72235 Cost_time: 1.06869s
I1112 12:18:23.811926 21056 train_model.py:209] =====>Epoch: 6679 D_loss: 102868.09375 G_loss: 0.90170 Ssim: 0.66528 Cost_time: 1.09463s
I1112 12:18:24.892297 21056 train_model.py:209] =====>Epoch: 6680 D_loss: 118568.27344 G_loss: 2.06072 Ssim: 0.77135 Cost_time: 1.07936s
I1112 12:18:25.973346 21056 train_model.py:209] =====>Epoch: 6681 D_loss: 40991.67969 G_loss: 2.10659 Ssim: 0.71723 Cost_time: 1.08005s
I1112 12:18:27.051629 21056 train_model.py:209] =====>Epoch: 6682 D_loss: 153677.87500 G_loss: 1.27149 Ssim: 0.80044 Cost_time: 1.07728s
I1112 12:18:28.111481 21056 train_model.py:209] =====>Epoch: 6683 D_loss: 119695.32031 G_loss: 2.99901 Ssim: 0.57306 Cost_time: 1.05684s
++++++++++++++++++++++++++++
from attentive-gan-derainnet.
I wonder if you can provide the tfevents file?
from attentive-gan-derainnet.
@hezw2016 ok I'will check it. The tfevents file will be uploaded later:)
from attentive-gan-derainnet.
我也遇到同样的问题,D_loss值很大,能达到十几万,请问这个问题怎么解决?是不是跟log函数的参数有0存在的原因,比如log(0)?
from attentive-gan-derainnet.
Related Issues (20)
- test_model.py中的label_path HOT 8
- 归一化层 HOT 2
- tf_io_pipline_tools模块引用错误
- 训练自己图片集有问题
- batch size HOT 2
- vgg16.npy HOT 1
- the bug HOT 3
- How to run this project at google colab ? HOT 1
- Colab Test HOT 2
- question about dataset
- Performance issues in /data_provider (by P3) HOT 1
- no file named "./data/vgg16.npy" HOT 4
- 训练时Discriminator的loss不收敛问题 HOT 4
- Question HOT 7
- no pretrained model "vgg16.npy"
- 關於訓練數據的問題(About train dataset size problem) HOT 1
- 测试的问题 HOT 1
- complexity
- 关于train_model.py文件无法正常运行 HOT 1
- loss function problem HOT 6
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 attentive-gan-derainnet.