Visible images access

[franckmarineai]

This is a great dataset but in order to fully test different models it would be good to also get access to the 'visible' images. I can't figure out how to get them. Thanks

[ssnirgudkar]

Hi,
Thanks. By visible images, do you mean optical images? If not, are you referring to the masked labeled images? If later, I think I can provide you these colored mask images.

[franckmarineai]

hi Shailesh Thanks for the quick turn around. I'm interested by the 'optical' images to train a GAN to convert 'optical' to 'infrared' images. I've already a model fully trained on the FLIR_ADAS dataset but it only includes cars, pedestrians, bycicles... and what I'm looking for is to train a model in the maritime space as we have ~100k opticle images we would like to convert to infrared in order to train an object detection model in the infrared spectrum. Regards,

…

________________________________ From: Shailesh Nirgudkar ***@***.***> Sent: Wednesday, January 25, 2023 1:08 PM To: uml-marine-robotics/MassMIND ***@***.***> Cc: Franck Barillaud ***@***.***>; Author ***@***.***> Subject: Re: [uml-marine-robotics/MassMIND] Visible images access (Issue #4) CAUTION: This email originated from outside of the organization. Do not click links or open attachments unless you recognize the sender and know the content is safe. Hi, Thanks. By visible images, do you mean optical images? If not, are you referring to the masked labeled images? If later, I think I can provide you these colored mask images. — Reply to this email directly, view it on GitHub<#4 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/A3CWKQXPIWOTKHRMJ3XTFWTWUEQUJANCNFSM6AAAAAAUGDJY7A>. You are receiving this because you authored the thread.Message ID: ***@***.***>

[ssnirgudkar]

Hi Frank,
If you are interested in (unlabeled) optical images in maritime domain, then I might be able to help you. Let me have a word with my advisor. Meanwhile, can you confirm that 1) you are interested in unlabeled optical images, 2) this is purely for research purpose and not for any commercial purpose?
Thanks.

[franckmarineai]

hi Shailesh, Yes I can confirm that we are interested by pairs (Optical/Infrared) unlabeled images for research purpose. Thanks for your help and support in this endeavor. Regards,

…

________________________________ From: Shailesh Nirgudkar ***@***.***> Sent: Thursday, January 26, 2023 3:15 AM To: uml-marine-robotics/MassMIND ***@***.***> Cc: Franck Barillaud ***@***.***>; Author ***@***.***> Subject: Re: [uml-marine-robotics/MassMIND] Visible images access (Issue #4) CAUTION: This email originated from outside of the organization. Do not click links or open attachments unless you recognize the sender and know the content is safe. Hi Frank, If you are interested in (unlabeled) optical images in maritime domain, then I might be able to help you. Let me have a word with my advisor. Meanwhile, can you confirm that 1) you are interested in unlabeled optical images, 2) this is purely for research purpose and not for any commercial purpose? Thanks. — Reply to this email directly, view it on GitHub<#4 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/A3CWKQUFYWECNMRSDVS7HC3WUHT3JANCNFSM6AAAAAAUGDJY7A>. You are receiving this because you authored the thread.Message ID: ***@***.***>

[ssnirgudkar]

Hi Frank,
When we recorded the dataset, the sensors (electro-optical, infrared and others) were mounted in different positions. I should add an image of the vehicle showing positions of the sensors. Anyway, the electro-optical and infrared cameras are not looking at the same scene exactly although there is some overlap. While creating this dataset, I did not consider electro-optical images because 1) they were not the subject of the research, 2) they did not produce anything during low visibility (dusk, night). So, unfortunately, I do not have paired images from optical and infrared sensors.

[franckmarineai]

Thanks Shailesh for the explanation. Unfortunately we need matching visible/infrared pairs of images so I'm afraid we can't use your dataset. Regards,

…

________________________________ From: Shailesh Nirgudkar ***@***.***> Sent: Thursday, January 26, 2023 5:23 PM To: uml-marine-robotics/MassMIND ***@***.***> Cc: Franck Barillaud ***@***.***>; Author ***@***.***> Subject: Re: [uml-marine-robotics/MassMIND] Visible images access (Issue #4) CAUTION: This email originated from outside of the organization. Do not click links or open attachments unless you recognize the sender and know the content is safe. Hi Frank, When we recorded the dataset, the sensors (electro-optical, infrared and others) were mounted in different positions. I should add an image of the vehicle showing positions of the sensors. Anyway, the electro-optical and infrared cameras are not looking at the same scene exactly although there is some overlap. While creating this dataset, I did not consider electro-optical images because 1) they were not the subject of the research, 2) they did not produce anything during low visibility (dusk, night). So, unfortunately, I do not have paired images from optical and infrared sensors. — Reply to this email directly, view it on GitHub<#4 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/A3CWKQTL6FMWHCRAINZ2LVTWUKXJXANCNFSM6AAAAAAUGDJY7A>. You are receiving this because you authored the thread.Message ID: ***@***.***>

[junghyun-avikus]

@ssnirgudkar
Hi, I am interested in the optical images as I am working on a smiliar task that frank has done.
In my case, the scenes do not necessarily have to be matched, but they need to overlap at least 60 ~ 70%.
Would it be the case? If so, can I ask for an acess to unlabeld optical images?

[ssnirgudkar]

@junghyun-avikus,
It is possible to have unlabeled optical images that overlap with LWIR images. However, it will require significant amount of work. Although the names of image files contain epoch, trying to find a file with closest epoch can be cumbersome.
May I know what is your affiliation and how will you be using these images?