LukasBommes/PV-Hawk

Add Flyr library to handle FLIR images

Resmung0 opened this issue · 4 comments

Resmung0 commented

Hi Lukas,

Congratulations on creating this incredible project! I learned a lot by reading your articles and your codes.

I have a lot of interest in photovoltaics and that's why I'm working on a project based on PV-Hawk that will be applied on the college campus where I graduated.

I realized that PV-Hawk needs thermal and optical images to be separated (I think you used the Flirpy library for this). However, I discovered a python library called Flyr which can extract thermographic and optical images from the original image taken by the FLIR thermal camera. I believe that adding this library to PV-Hawk stack is interesting not only for that, but also for many other features, such as:

  • Possibility of extracting and modifying the metadata responsible for creating thermograms (emissivity, distance to the object of interest, etc.);

  • Different ways of rendering the image (picture-in-picture, edge emphasis, and masking).

What do you thing about it?

I look forward to your reply!

LukasBommes commented

Hey Marcos,

thanks for this suggestion and also the one regarding the Kaggle dataset. We were actually thinking at one point to create a Kaggle challenge. However, I've already finished my PhD and went on to a new company and project. So, I do no have the time anymore to work on or support PV-Hawk.

Nontheless, it would be great if you want to fork the project and improve it. If at some point you want to merge back any improvements, I would be happy about that.

BTW: We have a partnering institute in Brazil which is using PV-Hawk. Maybe, I can connect you with them.

  • 👍1
Resmung0 commented

I understand, Lukas. I would be very grateful if you could introduce me to this institute here in Brazil.

But, just to make it clear: You don't have access to the module's annotation files, right? From the sample dataset that you shared here on Github.

LukasBommes commented

If you drop me an email (you can find my address here: https://www.hi-ern.de/profile/bommes_l) with your email address, then I will try to connect you with a post doc from our partner institute in Brazil.

Do you refer to the training dataset of the Mask R-CNN for PV module detection? This one is not public because it contains images from a variety of PV plants. To publish the dataset I would have to have gotten permission from the owners of all the plants. I know, it is quite a pity that I didn't put this extra effort to publish all my datasets also. But it would have been quite a lengthy procedure that I didn't really have the energy to pursue next to all the other obligations of the PhD.

Resmung0 commented

Regarding my request, I'm actually referring to the images you shared here in the PV-HAWK repository, the assets in releases. You have shared thermographic images of inspected PV modules using single and double rows (TIFF format) and RGB images in double rows (JPG format).

I would like to know if you could share the JSON files containing the module masks of the 3 datasets. So I can publish them on Kaggle.