darwinai / pl-covidnet-grad-cam Goto Github PK

Is your feature request related to a problem? Please describe.
Grad-cam implementations change based on the inspiration, the technology stack used, and the model used, even though we expect it to be model-agnostic. In general, it would be best if we can validate that grad-cam applied on COVID-Net is algorithmically and clinically correct.

Describe the bug
Currently, GradCam outputs 3 files, the original image, the mask, and the image after going through preprocessing (during which the dimensions of the image change and the image is cropped).

The mask is the same size as the preprocessed image, but we’d like to be able to apply the mask on the original image, so we should look to making the mask the same size, with what is used to compensate up to the discretion of the implementation (currently thinking to obstruct it like the unimportant areas in the mask).

To Reproduce
This behavior currently applies to all generated masks.

Expected behavior
The mask should be the same size of the original image, with what is used to compensate up to the discretion of the implementation (currently thinking to obstruct it like the unimportant areas in the mask).

Screenshots

Environment (please complete the following information):
N/A

Additional context
N/A

Describe the bug
The colors of the mask get inverted as well when we apply the inversion. This is because the mask and the underlying image is combined into one object through the canvas object and there was difficulty in applying the css effects used normally on the image without the mask on the individual images.

To Reproduce
This behavior applies in all cases.

Expected behavior
The colors of the mask must not get inverted when we apply the inversion.

Screenshots
If applicable, add screenshots to help explain your problem.

Environment (please complete the following information):
N/A

Additional context
N/A