search

yuxiangsun / RTFNet

RGB-Thermal Fusion Network for Semantic Segmentation of Urban Scenes
MIT License
166 stars 39 forks source link

visualize qualitative results #8

Closed DOHA-1 closed 4 years ago

DOHA-1 commented 4 years ago

first of all, thank you for sharing you code, i'm beginner in python and tensorflow and i don't know how to show the qualitative results ( predicted images) of testing, i just got quantitative values (probability of each class).

yuxiangsun commented 4 years ago

use: python3 run_demo.py

DOHA-1 commented 4 years ago

Thank you so much for your answer.

Le jeu. 14 nov. 2019 à 17:02, yuxiangsun notifications@github.com a écrit :

use: python3 run_demo.py

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/yuxiangsun/RTFNet/issues/8?email_source=notifications&email_token=AF43CN7LDVU2Y5HLV2PV4ELQTVZBXA5CNFSM4JKZO7AKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEECKYYY#issuecomment-553954403, or unsubscribe https://github.com/notifications/unsubscribe-auth/AF43CN7BH3IEVFD3IZ4UREDQTVZBXANCNFSM4JKZO7AA .

-- Doha Bouallal *PhD student - *Laboratoire IRF-SIC Ibn Zohr University - Agadir Tel: +212 6 63 08 82 91

DOHA-1 commented 4 years ago

Hi sir, i hope you are fine. sorry to bother you again, i have another question.

i want to train the RTFNet architecture with another dataset, how can i split the images and generate the train.txt , val.txt and test.txt ?

Le jeu. 14 nov. 2019 à 17:33, doha bouallal bouallaldoha@gmail.com a écrit :

Thank you so much for your answer.

Le jeu. 14 nov. 2019 à 17:02, yuxiangsun notifications@github.com a écrit :

use: python3 run_demo.py

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/yuxiangsun/RTFNet/issues/8?email_source=notifications&email_token=AF43CN7LDVU2Y5HLV2PV4ELQTVZBXA5CNFSM4JKZO7AKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEECKYYY#issuecomment-553954403, or unsubscribe https://github.com/notifications/unsubscribe-auth/AF43CN7BH3IEVFD3IZ4UREDQTVZBXANCNFSM4JKZO7AA .

-- Doha Bouallal *PhD student - *Laboratoire IRF-SIC Ibn Zohr University - Agadir Tel: +212 6 63 08 82 91

-- Doha Bouallal *PhD student - *Laboratoire IRF-SIC Ibn Zohr University - Agadir Tel: +212 6 63 08 82 91

yuxiangsun commented 4 years ago

You can randomly split your dataset to train 60%, val 20%, and test 20%.

DOHA-1 commented 4 years ago

Hi sir,

I want to thank you very much for taking the time to answer my beginner's questions. It's very kind of you. I hope I'm not disturbing you. I'm trying to understand how the RTFNet architecture works, in order to test it, and I didn't understand some details, which I will try to summarize as follows:

Does the network take as input the (registred image) fusion of the thermal and color image (which is in the folder " images") ? Or it takes the two images separately as input. for the display of these images I don't know why they are only visible when I have a black background, as shown in the following figure : [image: image.png] In the other hand, I noticed that the folder " labels" contains 2353 labels while the folder "images" contains 2390 images. do these labels correspond to these images? thank you for clarifying these points for me, I would be very grateful.

Have a good day.

Le ven. 22 nov. 2019 à 03:58, yuxiangsun notifications@github.com a écrit :

Hi sir, i hope you are fine. sorry to bother you again, i have another question. i want to train the RTFNet architecture with another dataset, how can i split the images and generate the train.txt , val.txt and test.txt ? Le jeu. 14 nov. 2019 à 17:33, doha bouallal bouallaldoha@gmail.com a écrit : … <#m-2858111388209495283> Thank you so much for your answer. Le jeu. 14 nov. 2019 à 17:02, yuxiangsun @.***> a écrit : > use: > python3 run_demo.py > > — > You are receiving this because you authored the thread. > Reply to this email directly, view it on GitHub > <#8 https://github.com/yuxiangsun/RTFNet/issues/8?email_source=notifications&email_token=AF43CN7LDVU2Y5HLV2PV4ELQTVZBXA5CNFSM4JKZO7AKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEECKYYY#issuecomment-553954403>,

or unsubscribe > https://github.com/notifications/unsubscribe-auth/AF43CN7BH3IEVFD3IZ4UREDQTVZBXANCNFSM4JKZO7AA . > -- Doha Bouallal *PhD student - Laboratoire IRF-SIC Ibn Zohr University - Agadir Tel: +212 6 63 08 82 91 -- Doha Bouallal PhD student - Laboratoire IRF-SIC Ibn Zohr University - Agadir Tel: +212 6 63 08 82 91 *

You can randomly split your dataset to train 60%, val 20%, and test 20%.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/yuxiangsun/RTFNet/issues/8?email_source=notifications&email_token=AF43CN2PD3CSCARJPTSFKRLQU5DHDA5CNFSM4JKZO7AKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEE4LYFY#issuecomment-557366295, or unsubscribe https://github.com/notifications/unsubscribe-auth/AF43CN7UFPMMGWBDNXCU6MLQU5DHDANCNFSM4JKZO7AA .

-- Doha Bouallal *PhD student - *Laboratoire IRF-SIC Ibn Zohr University - Agadir Tel: +212 6 63 08 82 91

yuxiangsun commented 4 years ago
  1. "images", not the separated, you can take a look at the data loader
  2. image cannot be seen
  3. some images at not labelled, images that are not labelled at not listed in the split txt files, so you no need to care about it
DOHA-1 commented 4 years ago

Thank you so much.

Le sam. 23 nov. 2019 à 04:02, yuxiangsun notifications@github.com a écrit :

  1. "images", not the separated, you can take a look at the data loader
  2. image cannot be seen
  3. some images at not labelled, images that are not labelled at not listed in the split txt files, so you no need to care about it

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/yuxiangsun/RTFNet/issues/8?email_source=notifications&email_token=AF43CNY7DAB33EOV6XQK76DQVCMK3A5CNFSM4JKZO7AKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEE7L2JI#issuecomment-557759781, or unsubscribe https://github.com/notifications/unsubscribe-auth/AF43CN65MXQGCRMFQ66SLGLQVCMK3ANCNFSM4JKZO7AA .

-- Doha Bouallal *PhD student - *Laboratoire IRF-SIC Ibn Zohr University - Agadir Tel: +212 6 63 08 82 91

DOHA-1 commented 4 years ago

Hello , I hope you and your family are doing well.

I have some questions about multispectral segmentation with RTFNet.

1- Before proceeding to the phase of concatenation of the thermal image and rgb, did you perform an alignment (registration) between the two? otherwise for my case I want to check quantitatively if the thermal and rgb images are aligned, are there any metrics to do that ?

2- My second question is : for my images I have only one class in the labels, while in yours you have 9 classes, what do I need to change in the code to adapt it to my case? should I add another class "background" or can it work with only one class?

I hope I don't bother you with my questions which can be trivial.

Thank you in advance and have a nice day. Stay safe.

Le dim. 24 nov. 2019 à 03:54, doha bouallal bouallaldoha@gmail.com a écrit :

Thank you so much.

Le sam. 23 nov. 2019 à 04:02, yuxiangsun notifications@github.com a écrit :

  1. "images", not the separated, you can take a look at the data loader
  2. image cannot be seen
  3. some images at not labelled, images that are not labelled at not listed in the split txt files, so you no need to care about it

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/yuxiangsun/RTFNet/issues/8?email_source=notifications&email_token=AF43CNY7DAB33EOV6XQK76DQVCMK3A5CNFSM4JKZO7AKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEE7L2JI#issuecomment-557759781, or unsubscribe https://github.com/notifications/unsubscribe-auth/AF43CN65MXQGCRMFQ66SLGLQVCMK3ANCNFSM4JKZO7AA .

-- Doha Bouallal *PhD student - *Laboratoire IRF-SIC Ibn Zohr University - Agadir Tel: +212 6 63 08 82 91

-- Doha Bouallal *PhD student - *Laboratoire IRF-SIC Ibn Zohr University - Agadir Tel: +212 6 63 08 82 91

yuxiangsun commented 4 years ago
  1. The registration is done in the MFNet paper. In addition, there are lots of literature discussing on this. You can google it.
  2. You need to change 9 in my code to 2. You have two classes: background and your foreground class.
DOHA-1 commented 4 years ago

Thank you very much for your answer.

Have a nice weekend.

Le sam. 2 mai 2020 à 08:34, yuxiangsun notifications@github.com a écrit :

  1. The registration is done in the MFNet paper. In addition, there are lots of literature discussing on this. You can google it.
  2. You need to change 9 in my code to 2. You have two classes: background and your foreground class.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/yuxiangsun/RTFNet/issues/8#issuecomment-622874709, or unsubscribe https://github.com/notifications/unsubscribe-auth/AF43CN7Q7T3ZSSUD7NSKK73RPPLINANCNFSM4JKZO7AA .

-- Doha Bouallal *PhD student - *Laboratoire IRF-SIC Ibn Zohr University - Agadir Tel: +212 6 63 08 82 91