Comments (4)
The bug record
from mxfont.
This is not a bug but a typical unstable GAN training.
We have also often observed similar unstable training as well.
Although the loss is explored from INDP_FACT
, we suspect that the unstable training is originated from "cross CE" which potentially increases the classifier weight values -- we couldn't test whether this is true due to the time limitation.
If you want to make your model more stable, then I suggest you remove the "cross CE" in the following lines:
- https://github.com/clovaai/mxfont/blob/main/trainer/fact_trainer.py#L250
- https://github.com/clovaai/mxfont/blob/main/trainer/fact_trainer.py#L265
It will slightly reduce your final performances but may make the training stable. But, please note that we also don't have exact clue why the unstable training happens.
(+ In practice, we select the "last model" under stable loss values)
from mxfont.
Hi, sorry for the inconvenience.
You can exclude the cross CE loss by setting the weight of cross CE loss to zero.
It can be set by giving --ac_cross_w argument when running train.py
like this:
python train.py cfgs/train.yaml --ac_cross_w 0.
from mxfont.
Closing the issue, assuming the answer resolves the problem.
Please re-open the issue as necessary.
from mxfont.
Related Issues (20)
- RuntimeError: indices should be either on cpu or on the same device as the indexed tensor (cpu) HOT 2
- Why do the models I trained predict the same results in two folders HOT 2
- There is a problem with the predicted character image HOT 3
- Got nothing after test using eval.py HOT 2
- Can the model output .ttf file? HOT 1
- OSError: symbolic link privilege not held HOT 1
- Why are there significant differences between the images generated in training and those generated in test files HOT 2
- dataset
- got a blank image during inference
- How can I export an onnx model HOT 4
- Is the content and style classifier pre trained, or are they trained together with mxfont HOT 6
- Training Set Consultation HOT 3
- font_indice font_indice HOT 3
- Consult "B.5. Training details" in the paper
- When reducing the training set characters, an error is reported: HOT 3
- The Role of Factorize and Defactorize HOT 2
- Hello, I have trained our model and found that the generation effect of characters with complex strokes is not very good. Do you have any good suggestions
- What are the meanings of style_facts_s style_facts_c, char_facts_s and char_facts_c HOT 5
- Classifier
- single Gpu report Bug
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 mxfont.