Comments (6)
Hi, thank you for your interest in our work,
To make sure I understand, you are training on all of the labels in a supervised manner, and the obtained mIoU is low at 40 epochs ?
Ill run the same on my end and I will get back to you.
from cct.
Yes, that's my setting. What's your supervised result with all the labeled data?
from cct.
Thanks,
If I remeber corretly, it was in the high seventies, but it was an earlier version of the code used for the paper, I didnt test it with this version, let me get back to you on this.
from cct.
Hi, I've just retrained for 35 out of 100 and gotten 73 miou, I expect to go a little bit higher before the end of training,
val_loss : 0.21543
Pixel_Accuracy : 0.937
Mean_IoU : 0.7310000061988831
I did actually find an error in the code that might lead to a lower number of epochs in your case, for the supervised mode, the number of iterations is calculated in terms of the unsupervised examples:
tbar = tqdm(range(len(self.supervised_loader)), ncols=135)
should be:
tbar = tqdm(range(len(self.supervised_loader)), ncols=135)
so in your case, the number of the unsupervised examples is lower (depending on number of examples you're training on), so this reduced the number of epochs, so for example, the result you're obtained for 40 epochs is, in reality, say only 20 epochs going through the whole labeled data, so you can simply train for longer and you'll get better results. I just pushed the correction for this bug, and if you run training for 100 epochs, I expect you'll get 74 / 75 at the end. And you can always gain a bit more in mIoU with semi mode or weakly mode.
I hope I answered your question.
Thanks.
from cct.
Thanks,
I suppose you mean
tbar = tqdm(range(len(self.unsupervised_loader)), ncols=135)
to
tbar = tqdm(range(len(self.supervised_loader)), ncols=135)
here.
I re-run the code and also get 73.4 mIoU at epoch 60.
from cct.
Yes exactly, thank you !
from cct.
Related Issues (20)
- Labels for unsupervised HOT 1
- Cusom dataset with one class HOT 2
- adjusting the model for input images with 4 channels HOT 3
- Training model only on foreground labels HOT 3
- inference with 4-channel model HOT 7
- Adjust the number of decoders HOT 2
- checkerboard HOT 7
- How can I train this model using my own dataset? HOT 1
- About checkpoint HOT 1
- About loss_unsup
- dataloader HOT 1
- Errors while inferencing the trained model HOT 1
- What are the required dimensions of `predict` and `target` for abCE loss? HOT 2
- Question about the paper "Semi-Supervised Semantic Segmentation with Cross-Consistency Training" HOT 1
- Program error during training HOT 1
- Training on custom datasets in CCT. HOT 4
- Question about the plot of Figure 2 in the paper HOT 1
- 自定义数据集问题
- Why use MSE instead of CE and KL divergence
- How to perform binary segmentation only ?
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 cct.