Skip to content
This repository has been archived by the owner on May 27, 2024. It is now read-only.

Question About .csv file #2

Open
zevyu opened this issue Dec 7, 2018 · 10 comments
Open

Question About .csv file #2

zevyu opened this issue Dec 7, 2018 · 10 comments

Comments

@zevyu
Copy link

zevyu commented Dec 7, 2018

I want to know how the csv file used to train mentor-dd is generated.
My understanding is to train the baseline model with a clean tag dataset and use the Corrupted Labels dataset to calculate the loss to get the csv file. Can you tell me the details of generating a csv file?

@roadjiang
Copy link
Contributor

We first train our model for 18 epochs on the noisy dataset. Then we use the model to evaluate on another small dataset, where we have some clean labels. The model will outputs all the feature (on the small dataset) to generate the csv.

@zevyu
Copy link
Author

zevyu commented Dec 8, 2018

I got it,thanks

@ruirui88
Copy link

We first train our model for 18 epochs on the noisy dataset. Then we use the model to evaluate on another small dataset, where we have some clean labels. The model will outputs all the feature (on the small dataset) to generate the csv.

Hi,I want to make sure how the csv file is generated.You said that it pre-trained model on the nosiy dataset firstly, and then evaluate the model on the small dataset(whose size is 10 percents?). So ,the clean labels in the csv file is the true labels of clean data, while the noisy labels is the prediction of the model?Is right?

@roadjiang
Copy link
Contributor

Details are in
https://github.com/google/mentornet/blob/master/TRAINING.md

clean label column: ground-truth labels on small clean dataset
noisy label column: given labels on the current noisy dataset
loss column: loss computed using the noisy label

@ruirui88
Copy link

Details are in
https://github.com/google/mentornet/blob/master/TRAINING.md

clean label column: ground-truth labels on small clean dataset
noisy label column: given labels on the current noisy dataset
loss column: loss computed using the noisy label

Sorry ,i don't quite get it. Whether if evaluating the pre-trained model on the clean and noisy dataset together? The samples whose ground-truth label and noisy label is the same comes from clean dataset, while the others come from noisy dataset. What's more, how does calculate the value in the clean label column for this noisy dataset. Is it manually annotated or prediciton of pre-trained model?

@wffzxyl
Copy link

wffzxyl commented May 13, 2019

Could you upload the files or code about the function 'provide_resnet_noisy_data' for extract resnet features in the cifa_eval.py(line 186)?

@wffzxyl
Copy link

wffzxyl commented May 13, 2019

Details are in
https://github.com/google/mentornet/blob/master/TRAINING.md
clean label column: ground-truth labels on small clean dataset
noisy label column: given labels on the current noisy dataset
loss column: loss computed using the noisy label

Sorry ,i don't quite get it. Whether if evaluating the pre-trained model on the clean and noisy dataset together? The samples whose ground-truth label and noisy label is the same comes from clean dataset, while the others come from noisy dataset. What's more, how does calculate the value in the clean label column for this noisy dataset. Is it manually annotated or prediciton of pre-trained model?

Have you finished the generation of the csv files? could you give me the csv file generation code. I can't found it in these files

@roadjiang
Copy link
Contributor

roadjiang commented May 13, 2019 via email

@AnnPe
Copy link

AnnPe commented Nov 12, 2019

We first train our model for 18 epochs on the noisy dataset. Then we use the model to evaluate on another small dataset, where we have some clean labels. The model will outputs all the feature (on the small dataset) to generate the csv.

Hi,I want to make sure how the csv file is generated.You said that it pre-trained model on the nosiy dataset firstly, and then evaluate the model on the small dataset(whose size is 10 percents?). So ,the clean labels in the csv file is the true labels of clean data, while the noisy labels is the prediction of the model?Is right?

Hi @ruirui88 , did you manage to create your csv file?

@AnnPe
Copy link

AnnPe commented Nov 12, 2019

Details are in
https://github.com/google/mentornet/blob/master/TRAINING.md
clean label column: ground-truth labels on small clean dataset
noisy label column: given labels on the current noisy dataset
loss column: loss computed using the noisy label

Sorry ,i don't quite get it. Whether if evaluating the pre-trained model on the clean and noisy dataset together? The samples whose ground-truth label and noisy label is the same comes from clean dataset, while the others come from noisy dataset. What's more, how does calculate the value in the clean label column for this noisy dataset. Is it manually annotated or prediciton of pre-trained model?

Have you finished the generation of the csv files? could you give me the csv file generation code. I can't found it in these files

Hi @wffzxyl , did you manage to generate the csv file? I can not reproduce the authors' results, so im afraid Im doing all the wrong way round

Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
None yet
Projects
None yet
Development

No branches or pull requests

5 participants