Comments (9)
Hi,
Could you please merely remove the ".zip" extension from the file named "2000" and check if it works properly? Without unzipping the file.
from ddad.
I do not find a checkpoints of feature_extractor named feat{DA_chp} in my downloaded files. Whether the fine-tuned model is not released, and should I fine-tune a feature_extractor myself?
I will release the checkpoints in the near future.
Prior to evaluation, it is necessary to fine-tune the feature extractor. To accomplish this, navigate to the config.yaml file and adjust the settings as outlined in the repository tables. Set DA_epochs and DA_chp based on the feature extractor epochs, and configure w with the corresponding value from the table.
After configuring the values in the config file, execute the following code to fine-tune the feature extractor. This process may take some time to complete:
Once the fine-tuning is done, you can proceed with model evaluation:
python main.py --domain_adaptation True
Now you can evaluate the model.
python main.py --eval True
If you wish to obtain information on misclassified samples and visualize reconstructions, you can customize the values of misclassifications and visualisation in the config file accordingly.
Please let me know if you face any other problems.
from ddad.
thank you! it works for me. However i encountered another problem when i load checkpoints of feature_extractor in 36 line of ddad.py (feature_extractor = domain_adaptation(self.unet, self.config, fine_tune=False) ). There is the notice as follow:
I do not find a checkpoints of feature_extractor named feat{DA_chp} in my downloaded files. Whether the fine-tuned model is not released, and should I fine-tune a feature_extractor myself?
from ddad.
I do not find a checkpoints of feature_extractor named feat{DA_chp} in my downloaded files. Whether the fine-tuned model is not released, and should I fine-tune a feature_extractor myself?
I will release the checkpoints in the near future.
Prior to evaluation, it is necessary to fine-tune the feature extractor. To accomplish this, navigate to the config.yaml file and adjust the settings as outlined in the repository tables. Set DA_epochs and DA_chp based on the feature extractor epochs, and configure w with the corresponding value from the table.
After configuring the values in the config file, execute the following code to fine-tune the feature extractor. This process may take some time to complete:
Once the fine-tuning is done, you can proceed with model evaluation: python main.py --domain_adaptation True
Now you can evaluate the model. python main.py --eval True
If you wish to obtain information on misclassified samples and visualize reconstructions, you can customize the values of misclassifications and visualisation in the config file accordingly. Please let me know if you face any other problems.
fine, thank you very much
from ddad.
Hi, i have fine-turn the feature extractor on hezelnut and screw. But the I-AUROC/P-AUROC is (96.4,99.3) on screw, and (99.8,98.2) on hazelnut. I set (load_chp=2000, DA_epochs=30, DA_chp=4, w=2 and w_DA=3) for screw and (load_chp=2000, DA_epochs=30, DA_chp=3, w=5 and w_DA=3) for hazelnut.
I set the corresponding parameters (w, and DA_chp) according to the readme, and set DA_epochs=30 and w_DA=3 on two categories. Is there any problem?
from ddad.
Apologies for the late response. I just cloned the code and conducted tests on both categories. Results are consistent with the reported ones. It's a bit challenging for me to pinpoint why you might be obtaining slightly different answers. Could you confirm that you're using the most recent version of our code?
To enhance usability, I plan to publish feature checkpoints over the weekend.
from ddad.
Also to decrease the fine-tuning time you can set the value of DA_epochs similar to DA_chp. It is not required to fine-tune for 30 epochs as it may be time-consuming.
from ddad.
Thank you very well. Sorry for not checking the issue for a long time. I certainty used most recent version. I think the only difference is batchsize(I set it as 12 because of limited GPU memory). I think it is the reason. Looking forward the checkpoints.
from ddad.
True, changing batch size will change the results. I have uploaded the MVTec checkpoints and will upload VisA checkpoints tomorrow.
from ddad.
Related Issues (20)
- Question about the initilization of "ch" in UNetModel.__init__() HOT 3
- I can not reproduce your results by using the uploaded checkpoints HOT 3
- How to set w_DA in Domain Adaptation? HOT 1
- Simple version of patchify()
- Can I train my own dataset?
- ModuleNotFoundError: No module named 'omegaconf'
- why am i not really training...... HOT 1
- Question about the precision HOT 3
- Question in making heat map HOT 2
- 為啥算異常評分不是用x0與x算 而是用y呢 怪怪唷
- img_size = 512 and change channel will report an error HOT 2
- Invalid load key HOT 3
- The AUROC value of the paper cannot be achieved using official checkpoints. HOT 6
- train HOT 2
- 我可以用自己的數據集嗎
- 我的檔案路徑可以用window寫法嗎
- The result cannot be achieved using official checkpoint. HOT 4
- 我用UBUNTU HOT 3
- results HOT 1
- 可以用3d數據集嗎 HOT 1
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 ddad.