search

lindawangg / COVID-Net

COVID-Net Open Source Initiative
Other
1.15k stars 480 forks source link

using your image from study gives false positive? #77

Closed jomollin closed 3 years ago

jomollin commented 4 years ago

Hallo, I used image A from your study, just a screen shot, and got it as covid-19 for 99.9 % Maybe you can send me the original image for testing

How that?

lindawangg commented 4 years ago

I'm not sure exactly which image that is but it's a non-COVID19 infection from this dataset: https://github.com/ieee8023/covid-chestxray-dataset. Having that image classified as covid-19 is totally possible as the models don't have a 100% positive predictive value.

jomollin commented 4 years ago

I'm checking additionally some images from https://github.com/agchung/Actualmed-COVID-chestxray-dataset which are marked with no finding, but always comes back to me as covid positive

Apekshagaonkar commented 4 years ago

Hello, Even I have the same issue. I tried to inference images from dataset you have used to train. https://www.kaggle.com/tawsifurrahman/covid19-radiography-database It shows covid positive to even normal and pneumonia case. Can you let us know if we have to do any preprocessing other than mentioned in your repository.

palatos commented 3 years ago

I'd like to second the request of the users above. I loaded the "COVIDNet-CXR3-C" and tried a bunch of normal (no illness) images from the datasets that go into COVIDx, and I'm always getting ~0.99 softmax output for the COVID-19 class. I'm not sure if it's a problem with loading the models or something with the preprocessing of the image.

It'd be nice to confirm.

haydengunraj commented 3 years ago

Closing this issue now. For any related problems with the more recent models, please open a new issue.