Comments (10)
Hi! The data is available here. I also added the link to the data to the README. Thanks for pointing this out!
Let me know if there are any further issues!
from detrtime.
Your prompt reply and willingness to share your data with me are greatly appreciated. Thank you very much for your kindness and generosity.
from detrtime.
Hi,I downloaded your dataset and noticed that its shape is [N, 500, 129]. I understand that the frequency is 500Hz, but I'm confused about why there are 129 channels when your paper mentions using a 128-channel EEG Geodesic Hydrocel system.
from detrtime.
That's a good question that needs clarification: during preprocessing, we compute the average of the signal at all 128 physical EEG electrodes and subtract it from the EEG signal at every electrode for every time point. This average of the 128 channels is then added as 129-th channel for every time point.
Let me know if you have any further questions!
from detrtime.
Hello,I want to konw what's your method of solving the probelm that the train set without any object on it.
Thanks
from detrtime.
Hi @Onlyknight ! Can you specify in more detail what your question is? I understand it as how we handle the case that during training, there might be no object in a sample.
If this is indeed your question: for the segmentation tasks we consider (eye events and sleep stages), this case does not occur, since there is always some eye event (fixation, saccade, blink) present (similar in the domain of sleep staging). The architecture can be generalized easily to have a "no object" class (like the original DETR for vision) in case you are processing time series data where "no object" is a valid case.
from detrtime.
Hi,@lu-wo ,thank you for your answer.
Yet,when I try to use the DETRTime for my training task on my dataset, I found that the train set must have object in order to finish the HungarianMatcher.
I hope to train the DETRTime on my dataset in order to detect the sleep apnea event.
But most samples of my train samples without any object , I mean that they are negative samples.
Now, I try to solve the problem with a method that I regard the normal time as a object (like the fixation in your eye event).But, I want to know how to set the biased_weights and empty_weight, can you give some guiding opinions?
Thanks.
from detrtime.
Hi @Onlyknight! It seems like you want to train a segmentation task with binary labels (sleep apnea event vs. none). The easiest way to do this, is simply assigning "no event" class 0, and "sleep apnea" class 1.
Let me know if you need further advice!
from detrtime.
Hello, I would like to ask, what is the basis for setting the loss weights and sampling weights in your network?
from detrtime.
Loss weights: The loss weights are taken from the original DETR implementation and can be adapted depending on the convergence of a particular loss component during training.
Sampling weights: You can set them to upsample minority classes. For example, in our ocular event datasets, fixation is the (temporal) majority class, while saccades and especially blinks occur less often. By setting the sampling weights, one can guarantee that the model is being updated oftentimes on the minority classes as well.
Let me know if you have further questions.
from detrtime.
Related Issues (3)
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 detrtime.