Problem about the bob.rppg code on HCI data

[Radim Špetlík]

Hi, I followed install instructions on "https://gitlab.idiap.ch/bob/bob.rppg.base" and I'm getting the following error:

ModuleNotFoundError: No module named 'bob.rppg.base.script.extract_boundingboxes'

[Guillaume Heusch]

Dear Radim,

Indeed, there is no script called extract_boundingboxes.py in bob/rppg/base, but I'm unable to reproduce the problem you encountered.

What I advise you to do is the following:

- install bob as per instructions given here: https://www.idiap.ch/software/bob/docs/bob/bob/stable/install.html

- After having activated your bob environment (i.e. source activate bob_py3), git clone the bob.rppg.base package: git clone g...@gitlab.idiap.ch:bob/bob.rppg.base.git

- go to the base folder, and do a buildout (i.e. cd bob.rppg.base; buildout)

Doing this way, it should build the package without problems, and you should be able to run the scripts. The documentation on the README here https://gitlab.idiap.ch/bob/bob.rppg.base is indeed a bit outdated, and I'll correct it asap.

Note also that you may need to download data "required" by this package: either the Manhob HCI databse or the COHFACE database. Data are not provided with it, and you should download them by yourself

I hope this will help, and let me know how it goes

Best,

Guillaume

[Radim Špetlík]

Hi,

I don't know what happened, but I must have some weird stable version for python 2 - it seems that the scripts contain python2 syntax, but the environment is python3. I followed the instructions at the site you mention. What could have happened?

--
-- You received this message because you are subscribed to the Google Groups bob-devel group. To post to this group, send email to bob-...@googlegroups.com. To unsubscribe from this group, send email to bob-devel+unsubscribe@googlegroups.com. For more options, visit this group at https://groups.google.com/d/forum/bob-devel or directly the project website at http://idiap.github.com/bob/
---
You received this message because you are subscribed to a topic in the Google Groups "bob-devel" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/bob-devel/aTuPFHT5zDk/unsubscribe.
To unsubscribe from this group and all its topics, send an email to bob-devel+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

--

Radim V. Špetlík

Gagarinova 791 / 3, Liberec 7, 460 07

GSM: 00420 775 277 208

Skype ID: radim.spetlik

O.A.D.G.

[Guillaume Heusch]

Hi Radim,

Well, to be honest, it is hard to guess what could have happened. 

The package was indeed written with python2 syntax, but should works in a python3 environment (the one you created).

So, did you manage to make it work ?

Best,

Guillaume

Le lundi 26 février 2018 09:40:31 UTC+1, Radim Špetlík a écrit :

Hi,

I don't know what happened, but I must have some weird stable version for python 2 - it seems that the scripts contain python2 syntax, but the environment is python3. I followed the instructions at the site you mention. What could have happened?

On 26 February 2018 at 09:36, Guillaume Heusch <guillaum...@gmail.com> wrote:

Dear Radim,

Indeed, there is no script called extract_boundingboxes.py in bob/rppg/base, but I'm unable to reproduce the problem you encountered.

What I advise you to do is the following:

- install bob as per instructions given here: https://www.idiap.ch/software/bob/docs/bob/bob/stable/install.html

- After having activated your bob environment (i.e. source activate bob_py3), git clone the bob.rppg.base package: git clone g...@gitlab.idiap.ch:bob/bob.rppg.base.git

- go to the base folder, and do a buildout (i.e. cd bob.rppg.base; buildout)

Doing this way, it should build the package without problems, and you should be able to run the scripts. The documentation on the README here https://gitlab.idiap.ch/bob/bob.rppg.base is indeed a bit outdated, and I'll correct it asap.

Note also that you may need to download data "required" by this package: either the Manhob HCI databse or the COHFACE database. Data are not provided with it, and you should download them by yourself

I hope this will help, and let me know how it goes

Best,

Guillaume

Le samedi 24 février 2018 15:35:32 UTC+1, Radim Špetlík a écrit :

Hi, I followed install instructions on "https://gitlab.idiap.ch/bob/bob.rppg.base" and I'm getting the following error:

ModuleNotFoundError: No module named 'bob.rppg.base.script.extract_boundingboxes'

--
-- You received this message because you are subscribed to the Google Groups bob-devel group. To post to this group, send email to bob-...@googlegroups.com. To unsubscribe from this group, send email to bob-devel+...@googlegroups.com. For more options, visit this group at https://groups.google.com/d/forum/bob-devel or directly the project website at http://idiap.github.com/bob/

---
You received this message because you are subscribed to a topic in the Google Groups "bob-devel" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/bob-devel/aTuPFHT5zDk/unsubscribe.

To unsubscribe from this group and all its topics, send an email to bob-devel+...@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

[Radim Špetlík]

Yes, I did. I just wanted to let you know. Find&replace of the python3 syntax errors was needed...

Anyway, I'm having difficulties reproducing the results in the case of li-hci-cvpr14.py script. I'm getting a higher RMSE and a lower Pearson's correlation coefficient though it looks like the scripts executed successfully... 

The hci database was downloaded directly from the mahnob-db site, the https://gitlab.idiap.ch/bob/bob.rppg.base master branch was used.

--
-- You received this message because you are subscribed to the Google Groups bob-devel group. To post to this group, send email to bob-...@googlegroups.com. To unsubscribe from this group, send email to bob-devel+unsubscribe@googlegroups.com. For more options, visit this group at https://groups.google.com/d/forum/bob-devel or directly the project website at http://idiap.github.com/bob/

---
You received this message because you are subscribed to a topic in the Google Groups "bob-devel" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/bob-devel/aTuPFHT5zDk/unsubscribe.

To unsubscribe from this group and all its topics, send an email to bob-devel+unsubscribe@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

[Guillaume Heusch]

Glad you managed to make it work.

Regarding the results, are you far from what is published in the paper ?

[Radim Špetlík]

Not so far, the numbers are:

[INFO] Root Mean Squared Error = 9.81

[INFO] Mean of error-rate percentage = 0.10

[INFO] Pearson's correlation = 0.54 (0.00 significance)

The published:

RMSE = 8.12

Person's = 0.7

But if I understand the procedures correctly, the numbers should be exactly the same - I am using only the codes provided by you. I assume that during the installation the dependencies are handled so the setup is exactly the same as you had when computed the results...

--
-- You received this message because you are subscribed to the Google Groups bob-devel group. To post to this group, send email to bob-...@googlegroups.com. To unsubscribe from this group, send email to bob-devel+unsubscribe@googlegroups.com. For more options, visit this group at https://groups.google.com/d/forum/bob-devel or directly the project website at http://idiap.github.com/bob/

---
You received this message because you are subscribed to a topic in the Google Groups "bob-devel" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/bob-devel/aTuPFHT5zDk/unsubscribe.

To unsubscribe from this group and all its topics, send an email to bob-devel+unsubscribe@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

[Guillaume Heusch]

Well, It has been a while since I ran these experiments, and although we are doing our best to make them reproducible, our codes are kind of in a constant evolution ... 

I'll run the script within my current framework and I'll let you know on the outcome.

Best,

Guillaume

[Radim Špetlík]

Great, thank you very much...

--
-- You received this message because you are subscribed to the Google Groups bob-devel group. To post to this group, send email to bob-...@googlegroups.com. To unsubscribe from this group, send email to bob-devel+unsubscribe@googlegroups.com. For more options, visit this group at https://groups.google.com/d/forum/bob-devel or directly the project website at http://idiap.github.com/bob/

---
You received this message because you are subscribed to a topic in the Google Groups "bob-devel" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/bob-devel/aTuPFHT5zDk/unsubscribe.

To unsubscribe from this group and all its topics, send an email to bob-devel+unsubscribe@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

[Radim Špetlík]

Dear Guillaume, I forgot to mention that I followed also this site: "https://www.idiap.ch/software/bob/docs/bob/bob.rppg.base/master/guide_bbox.html" 

... 

[Guillaume Heusch]

Hi Radim,

I just finished re-running the experiments, and I get the same results as published. Here's the output of the last script:

[INFO] ==================

[INFO] === STATISTICS ===

[INFO] Root Mean Squared Error = 8.12

[INFO] Mean of error-rate percentage = 0.08

[INFO] Pearson's correlation = 0.70 (0.00 significance)

But I figured that I had to correct some stuff as compared to what is in 'li-hci-cvpr.py', and it concerns the bounding boxes. In the extract_signals.py script as it is done now, both bounding boxes and keypoints for the first frame of each video sequences are automatically loaded from the database metadata (and you should not provide a bounding boxes directory anymore). Metadata are loaded from the bob.db.hci_tagging package directly.

So basically, you should run the following command to extract signals:

./bin/cvpr14_extract_signals.py hci --protocol cvpr14 --dbdir hci --facedir experiments/paper/li-hci-cvpr14/face --bgdir experiments/paper/li-hci-cvpr14/bg --npoints 200 --indent 10 -v

Could you try and let me know how it goes ?

Best,

Guillaume

[Radim Špetlík]

Dear Guillaume,

I think I found the problem. Some mahnob files are garbled after extraction, seems like the first "key" frame (intra frame) is missing. Could you tell me what version of ffmpeg and codecs are you running please?

--
-- You received this message because you are subscribed to the Google Groups bob-devel group. To post to this group, send email to bob-...@googlegroups.com. To unsubscribe from this group, send email to bob-devel+unsubscribe@googlegroups.com. For more options, visit this group at https://groups.google.com/d/forum/bob-devel or directly the project website at http://idiap.github.com/bob/

---
You received this message because you are subscribed to a topic in the Google Groups "bob-devel" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/bob-devel/aTuPFHT5zDk/unsubscribe.

To unsubscribe from this group and all its topics, send an email to bob-devel+unsubscribe@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

[Guillaume Heusch]

Hi Radim,

I actually use bob.io.video to load the original videos ... So I don't know exactly what's under the hood there.

However, you can have a look at its documentation here: https://www.idiap.ch/software/bob/docs/bob/bob.io.video/master/video.html

Hope it helps,

Best,

Guillaume

[Radim Špetlík]

Dear Guillaume,

I already took a look and that is why I ask about the FFmpeg version :) Or better, could you check if the extraction of the signal from the videos is made using FFmpeg or libav backend, please? If so, I would also appreciate a version of the used library...

I also have another question. Do you have the COHFACE videos in a raw format? There is a paper by McDuff et al. http://alumni.media.mit.edu/~djmcduff/assets/publications/McDuff_2017_Impact2.pdf where it is shown that a video compression has severe effects on the PPG signal reconstruction. In my opinion, the COHFACE database needs either be distributed uncompressed or with experiments showing how the reconstructed signal changes under various compression settings. Would you consider performing such experiments?

Also, is there any way to speed up the procedure of granting access to the database? I requested the access week ago and nothing happened since.

--
-- You received this message because you are subscribed to the Google Groups bob-devel group. To post to this group, send email to bob-...@googlegroups.com. To unsubscribe from this group, send email to bob-devel+unsubscribe@googlegroups.com. For more options, visit this group at https://groups.google.com/d/forum/bob-devel or directly the project website at http://idiap.github.com/bob/

---
You received this message because you are subscribed to a topic in the Google Groups "bob-devel" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/bob-devel/aTuPFHT5zDk/unsubscribe.

To unsubscribe from this group and all its topics, send an email to bob-devel+unsubscribe@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.