MVE using Machine Video Editor to correct Face alignments and Landmarks - RE-EXTRACT

[angeloshredder]

I also get the " no faces detected error "

I think
The tutorial uses DFL's aligned image.
I don't know how to start extracting from the video.
Still trying

yes , i used already extracted faces ( dfl extraction) , it's possible to do extraction by using MVE,  but,  because  the objective was just to correct the bad extracted ( also not extracted faces ) , so i used DFL extraction.  
so, as i said before the objective is to correct some bad aligned faces and then export the corrected alignments as DFL images

 

[siao]

I also get the " no faces detected error "

I think
The tutorial uses DFL's aligned image.
I don't know how to start extracting from the video.
Still trying

yes , i used already extracted faces ( dfl extraction) , it's possible to do extraction by using MVE,  but,  because  the objective was just to correct the bad extracted ( also not extracted faces ) , so i used DFL extraction.  
so, as i said before the objective is to correct some bad aligned faces and then export the corrected alignments as DFL images

Yes

That's one of the reasons I learned to use

---

I was using aligned data and not trying to  extract from video. I followed the steps directly as shown in the video, but it says Faces not detected.

Maybe the update is incomplete
Need time to update
I spent a long time updating

Do you get a GPU option?

 

[Roy_B]

Same problem here. I have a GTX 1060 6GB but i get "no faces found" error. When i try to install dependencies i get the thessage "impossible to find specified path".


EDIT: all problems seems related to unicode characters in my install path (i used "!MVE" as my install folder). Thanks SPT for the help


I've made a batch script as requested by SPT to manage bad aligned images and use them into MVE. Just copy/paste the following text and save it as *.bat file, then run it from DFL root folder. First make backup of your facesets first just for precaution...i've got to do more tests

EDIT: assuming that you have first checked and deleted all bad DFL landmarks images from SRC or DST "aligned_debug" folder to recycle bin, follow on screen instructions.
The script will create a folder called "MVE Project" inside DFL main install folder, with the sub-folders "aligned" and "FRAMES" (like in this  made by angeloshredder) and their relative extracted frames and faces.
Note: the "MVE Project" folder will contain only the necessary files that DFL can't handle correctly, ready to be fixed with MVE:

 OFF

CLS

ECHO ==================================================================================
ECHO                   RECOVER ALIGNED_DEBUG IMAGES FROM RECYCLE BIN:
ECHO ==================================================================================
ECHO.
ECHO     Move deleted aligned_debug images from recycle bin directly into the folder
ECHO     "recovered_debug" of your SRC/DST dataset. DON'T RECOVER TO ORIGINAL location
ECHO.
ECHO ==================================================================================
ECHO.

SET /P C=DO YOU WANT TO RECOVER SOURCE OR DESTINATION DEBUG IMAGES [S/D, Q to ABORT]?
IF /I "%C%" EQU "S" GOTO :SRC
IF /I "%C%" EQU "D" GOTO :DST
IF /I "%C%" EQU "Q" GOTO :ABORT
GOTO :CHOICE

:SRC
MD "workspace\data_src\recovered_debug"
START shell:RecycleBinFolder
GOTO :CONTINUE

:DST
MD "workspace\data_dst\recovered_debug"
START shell:RecycleBinFolder
GOTO :CONTINUE

:ABORT
GOTO :EOF



:CONTINUE

ECHO.
ECHO.
ECHO.
ECHO ######################## PROCEED WITH NEXT STEP WHEN DONE ########################
ECHO.
ECHO.
ECHO.
PAUSE
ECHO.
ECHO.
ECHO.
ECHO ==================================================================================
ECHO        PREPARING IMAGE DATASET WITH BAD ALIGNMENTS FOR MACHINE VIDEO EDITOR:
ECHO ==================================================================================
ECHO.
ECHO       Before proceeding, first remember to restore original filename of your
ECHO           SRC or DST extracted faces by using the dedicated DFL utility,
ECHO                   otherwise the script will NOT work correctly!
ECHO.
ECHO             WHEN DONE, SELECT SOURCE or DESTINATION DATASET TO PROCESS
ECHO.
ECHO ==================================================================================
ECHO.
ECHO.



SET /P C=DO YOU WANT TO PROCESS SOURCE OR DESTINATION DATASET [S/D, Q to ABORT]?
IF /I "%C%" EQU "S" GOTO :SOURCE
IF /I "%C%" EQU "D" GOTO :DESTINATION
IF /I "%C%" EQU "Q" GOTO :ABORT
GOTO :CHOICE



:SOURCE

MD "MVE Project\data_src\FRAMES"

XCOPY /I /K /Q /Y "workspace\data_src\*" "MVE Project\data_src\FRAMES"
ATTRIB +R "MVE Project\data_src\FRAMES\*"

XCOPY /I /K /Q /Y /R "workspace\data_src\recovered_debug" "MVE Project\data_src\FRAMES"

DEL /AR /F /Q "MVE Project\data_src\FRAMES"
XCOPY /U /I /K /Q /Y "workspace\data_src\*" "MVE Project\data_src\FRAMES"



MD "MVE Project\data_src\aligned"

XCOPY /I /K /Q /Y "workspace\data_src\aligned\*" "MVE Project\data_src\aligned"
ATTRIB +R "MVE Project\data_src\aligned\*"

REN "workspace\data_src\recovered_debug\*.jpg" *.
REN "workspace\data_src\recovered_debug\*" *_0.jpg

XCOPY /I /K /Q /Y /R "workspace\data_src\recovered_debug\*" "MVE Project\data_src\aligned"

DEL /AR /F /Q "MVE Project\data_src\aligned"
XCOPY /U /I /K /Q /Y "workspace\data_src\aligned\*.jpg" "MVE Project\data_src\aligned"



SETLOCAL ENABLEDELAYEDEXPANSION
PUSHD WORKSPACE\DATA_SRC\RECOVERED_DEBUG
FOR %%A IN (*_*) DO ( SET FILE=%%A
REN "!FILE!" "!FILE:_0=!"
)
POPD



ECHO.
ECHO ==================================================================================
ECHO  COPIED ALL NECESSARY FILES INTO "MVE PROJECT" FOLDER INSIDE DFL INSTALL location
ECHO ==================================================================================
ECHO.
ECHO.
ECHO.
ECHO DONE
PAUSE



:DESTINATION

MD "MVE Project\data_dst\FRAMES"

XCOPY /I /K /Q /Y "workspace\data_dst\*" "MVE Project\data_dst\FRAMES"
ATTRIB +R "MVE Project\data_dst\FRAMES\*"

XCOPY /I /K /Q /Y /R "workspace\data_dst\recovered_debug" "MVE Project\data_dst\FRAMES"

DEL /AR /F /Q "MVE Project\data_dst\FRAMES"
XCOPY /U /I /K /Q /Y "workspace\data_dst\*" "MVE Project\data_dst\FRAMES"



MD "MVE Project\data_dst\aligned"

XCOPY /I /K /Q /Y "workspace\data_dst\aligned\*" "MVE Project\data_dst\aligned"
ATTRIB +R "MVE Project\data_dst\aligned\*"

REN "workspace\data_dst\recovered_debug\*.jpg" *.
REN "workspace\data_dst\recovered_debug\*" *_0.jpg

XCOPY /I /K /Q /Y /R "workspace\data_dst\recovered_debug\*" "MVE Project\data_dst\aligned"

DEL /AR /F /Q "MVE Project\data_dst\aligned"
XCOPY /U /I /K /Q /Y "workspace\data_dst\aligned\*.jpg" "MVE Project\data_dst\aligned"



SETLOCAL ENABLEDELAYEDEXPANSION
PUSHD WORKSPACE\DATA_DST\RECOVERED_DEBUG
FOR %%A IN (*_*) DO ( SET FILE=%%A
REN "!FILE!" "!FILE:_0=!"
)
POPD



ECHO.
ECHO ==================================================================================
ECHO  COPIED ALL NECESSARY FILES INTO "MVE PROJECT" FOLDER INSIDE DFL INSTALL location
ECHO ==================================================================================
ECHO.
ECHO.
ECHO.
ECHO DONE
PAUSE





:ABORT
GOTO :EOF

 

[angeloshredder]

Same problem here. I have a GTX 1060 6GB but i get "no faces found" error. When i try to install dependencies i get the thessage "impossible to find specified path".


EDIT: all problems seems related to unicode characters in my install path (i used "!MVE" as my install folder). Thanks SPT for the help


I've made a batch script as requested by SPT to manage bad aligned images and use them into MVE. Just copy/paste the following text and save it as *.bat file, then run it from DFL root folder. First make backup of your facesets first just for precaution...i've got to do more tests

EDIT: assuming that you have first checked and deleted all bad DFL landmarks images from SRC or DST "aligned_debug" folder to recycle bin, follow on screen instructions.
The script will create a folder called "MVE Project" inside DFL main install folder, with the sub-folders "aligned" and "FRAMES" (like in this  made by angeloshredder) and their relative extracted frames and faces.
Note: the "MVE Project" folder will contain only the necessary files that DFL can't handle correctly, ready to be fixed with MVE:

ECHO OFF

CLS

START shell:RecycleBinFolder



ECHO ==================================================================================
ECHO      PREPARING .JPG IMAGE DATASET WITH BAD LANDMARKS FOR MACHINE VIDEO EDITOR
ECHO    Move deleted landmark debug images from recycle bin directly into the folder
ECHO "recovered_landmarks" of your DST/SRC dataset. DON'T RECOVER TO ORIGINAL location!
ECHO.
ECHO             WHEN DONE, SELECT SOURCE or DESTINATION DATASET TO PROCESS
ECHO ==================================================================================



SET /P C=DO YOU WANT TO PROCESS SOURCE OR DESTINATION DATASET [S/D, Q to ABORT]?
IF /I "%C%" EQU "S" GOTO :SOURCE
IF /I "%C%" EQU "D" GOTO :DESTINATION
IF /I "%C%" EQU "Q" GOTO :ABORT
GOTO :CHOICE



:SOURCE
MD "workspace\data_src\recovered_debug"
MD "MVE Project\FRAMES"
MD "MVE Project\aligned"

XCOPY /I /K /Q /Y "workspace\data_src\*.jpg" "MVE Project\FRAMES"
XCOPY /I /K /Q /Y "workspace\data_src\aligned\*.jpg" "MVE Project\aligned"

ATTRIB +R "workspace\data_src\recovered_debug\*.jpg"
XCOPY /I /K /Q /Y "workspace\data_src\recovered_debug" "MVE Project\FRAMES"
DEL /AR /Q "MVE Project\FRAMES"

ATTRIB -R "workspace\data_src\recovered_debug\*.jpg"
REN "workspace\data_src\recovered_debug\*.jpg" *.
REN "workspace\data_src\recovered_debug\*" *_0.jpg

ATTRIB +R "workspace\data_src\recovered_debug\*.jpg"
XCOPY /I /K /Q /Y "workspace\data_src\recovered_debug" "MVE Project\aligned"
DEL /AR /Q "MVE Project\aligned"
ATTRIB -R "workspace\data_src\recovered_debug\*.jpg"



:DESTINATION
MD "workspace\data_dst\recovered_debug"
MD "MVE Project\FRAMES"
MD "MVE Project\aligned"

XCOPY /I /K /Q /Y "workspace\data_dst\*.jpg" "MVE Project\FRAMES"
XCOPY /I /K /Q /Y "workspace\data_dst\aligned\*.jpg" "MVE Project\aligned"

ATTRIB +R "workspace\data_dst\recovered_debug\*.jpg"
XCOPY /I /K /Q /Y "workspace\data_dst\recovered_debug" "MVE Project\FRAMES"
DEL /AR /Q "MVE Project\FRAMES"

ATTRIB -R "workspace\data_dst\recovered_debug\*.jpg"
REN "workspace\data_dst\recovered_debug\*.jpg" *.
REN "workspace\data_dst\recovered_debug\*" *_0.jpg

ATTRIB +R "workspace\data_dst\recovered_debug\*.jpg"
XCOPY /I /K /Q /Y "workspace\data_dst\recovered_debug" "MVE Project\aligned"
DEL /AR /Q "MVE Project\aligned"
ATTRIB -R "workspace\data_dst\recovered_debug\*.jpg"



:ABORT
GOTO :EOF



ECHO ON

Hi, Roy_B 
Do we need to empty recycle-bin before execution and before deleting the unwanted aligned_debugs  ?
after execution it shows the recycle-bin and gives me this :

and there is no Folder named "recovred_landmarks" created neither in src nor in dst , do i need to create it manualy ?

 

[Roy_B]

"angeloshredder"

I've corrected the script (sorry for the mistake, it should be correct this time) and no, you don't have to empty recycle bin first (anyway it's better to empty it first, just to avoid confusion if you have many files into it).

After starting the script and before selecting SRC or DST the, use the new explorer "recycle bin" window to manually drag (or cut/paste) the necessary files into "recovered_debug" folder inside your DFL data_src folder (or data_dst, depending on your choice)

After you have selected SRC or DST the script will make a copy of needed files to be edited in MVE into a newly created folder called "MVE Project", located inside your DFL main folder: just the previously deleted "mis-aligned" faces with their relative extracted video frames will be copied,

This script is intended for those situations where you are dealing with those "impossible" faces to align with DFL's manual re-extract.

 

[siao]

"angeloshredder"

I've corrected the script (sorry for the mistake, it should be correct this time) and no, you don't have to empty recycle bin first (anyway it's better to empty it first, just to avoid confusion if you have many files into it).

After starting the script and before selecting SRC or DST the, use the new explorer "recycle bin" window to manually drag (or cut/paste) the necessary files into "recovered_debug" folder inside your DFL data_src folder (or data_dst, depending on your choice)

After you have selected SRC or DST the script will make a copy of needed files to be edited in MVE into a newly created folder called "MVE Project", located inside your DFL main folder:  just the previously deleted "mis-aligned" faces with their relative extracted video frames will be copied,

This script is intended for those situations where you are dealing with those "impossible" faces to align with DFL's manual re-extract.

If the data_dst picture is *.png
Could not copy to "FRAMES"
Please check
Thank you

 

[Roy_B]

If the data_dst picture is *.png
Could not copy to "FRAMES"
Please check
Thank you

 
As the script title suggests, it is intended for JPG files. If you want to use it with PNG files, all you have to do is replace the ".jpg" part of the script text with ".png" (or ".*" but it will copy any other file types) and that's it

 

[siao]

If the data_dst picture is *.png
Could not copy to "FRAMES"
Please check
Thank you

 
As the script title suggests, it is intended for JPG files. If you want to use it with PNG files, all you have to do is replace the ".jpg" part of the script text with ".png" (or ".*" but it will copy any other file types) and that's it

Yes.
Sorry i missed that

another thing
Can I keep the deleted pictures?

I modify the script
"aligned" keep deleted pictures
but
"FRAMES" still keep all pictures

I think
If the folder is large
Can shorten data loading time
And easy to find

English is not good
Hope you understand what i mean
Thank you

 

[Roy_B]

This script doesn't delete anything, it copies only the files needed to process into MVE (as in the video tutorial by angeloshredder). Nothing more nothing less.

PNG image format is pretty useless imho, you'd better use JPG format for video frames exraction and get the very same results as PNG: in DFL you will get almost the same number of "bad aligned faces" to correct, no matter the image format you use. Same thing for the final video result quality

 

[siao]

This script doesn't delete anything, it copies only the files needed to process into MVE (as in the video tutorial by angeloshredder). Nothing more nothing less.

PNG image format is pretty useless imho, you'd better use JPG format for video frames exraction and get the very same results as PNG: in DFL you will get almost the same number of "bad aligned faces" to correct, no matter the image format you use. Same thing for the final video result quality

Yes I know
But that is to copy the good alignment picture to MVE
I mean
I want to copy the bad alignment picture to MVE
And the same number of data_dst pictures

I modify the script
I can copy the bad alignment image to MVE
But the FRAMES folder copied all the data_dst pictures

I am not a professional
Don't understand how to modify the script correctly

--------
While
png or jpg change as I like


it's okay
Your script is good
Almost the same as I thought
Thank you

 

[Roy_B]

"SPT" "angeloshredder" "siao"
Now that i have been able to do proper tests, the script should work fine without problems (even for both JPG / PNG image formats). I edited my previous post, please refer to that.

 

[SPT]

"SPT"  "angeloshredder" "siao"
Now that i have been able to do proper tests, the script should work fine without problems (even for both JPG / PNG image formats). I edited my previous post, please refer to that.

So, I still didn't have time to try it yet, but just to be clear, the workflow should be :

Extract dst, sort & delete wrong frames, recover original file name, removing "_0" from the names, make a copy of aligned_debug folder, paste in it everything from dst_aligned, delete these pictures, now delete the pictures with no faces to re-extract, copy what's left to the real aligned_debug, delate, then use MANUAL RE-EXTRACT.

Then use your script, and it will create a folder with original full frames from the video corresponding to the frames Manual Re extract didn't manage to create. Correct ?

 

[siao]

"SPT"  "angeloshredder" "siao"
Now that i have been able to do proper tests, the script should work fine without problems (even for both JPG / PNG image formats). I edited my previous post, please refer to that.

Hello
The script is very good now
Thank you
Only one question
In the FRAMES folder
It should be the corresponding data_dst picture
But now it is the deleted aligned_debug picture
Did I understand it wrong?

---

"SPT"  "angeloshredder" "siao"
Now that i have been able to do proper tests, the script should work fine without problems (even for both JPG / PNG image formats). I edited my previous post, please refer to that.

So, I still didn't have time to try it yet, but just to be clear, the workflow should be :

Extract dst, sort & delete wrong frames, recover original file name, removing "_0" from the names, make a copy of aligned_debug folder, paste in it everything from dst_aligned, delete these pictures, now delete the pictures with no faces to re-extract, copy what's left to the real aligned_debug, delate, then use MANUAL RE-EXTRACT.

Then use your script, and it will create a folder with original full frames from the video corresponding to the frames Manual Re extract didn't manage to create. Correct ?

Hello
My understanding is
All processes must be completed in DFL first
Then "faceset MANUAL RE-EXTRACT DELETED ALIGNED_DEBUG"
Move to MVE correction

You can refer to the previous instructions
//deep.whitecatchel.ru/literotica/forums/thre...s-and-landmarks-re-extract?pid=30890#pid30890

 

[SPT]

"SPT"  "angeloshredder" "siao"
Now that i have been able to do proper tests, the script should work fine without problems (even for both JPG / PNG image formats). I edited my previous post, please refer to that.

Hello
The script is very good now
Thank you
Only one question
In the FRAMES folder
It should be the corresponding data_dst picture
But now it is the deleted aligned_debug picture
Did I understand it wrong?

---

"SPT"  "angeloshredder" "siao"
Now that i have been able to do proper tests, the script should work fine without problems (even for both JPG / PNG image formats). I edited my previous post, please refer to that.

So, I still didn't have time to try it yet, but just to be clear, the workflow should be :

Extract dst, sort & delete wrong frames, recover original file name, removing "_0" from the names, make a copy of aligned_debug folder, paste in it everything from dst_aligned, delete these pictures, now delete the pictures with no faces to re-extract, copy what's left to the real aligned_debug, delete, then use MANUAL RE-EXTRACT.

Then use your script, and it will create a folder with original full frames from the video corresponding to the frames Manual Re extract didn't manage to create. Correct ?

Hello
My understanding is
All processes must be completed in DFL first
Then "faceset MANUAL RE-EXTRACT DELETED ALIGNED_DEBUG"
Move to MVE correction

You can refer to the previous instructions
//deep.whitecatchel.ru/literotica/forums/thre...s-and-landmarks-re-extract?pid=30890#pid30890

---

If you get the pics from debug either there's still something wrong with the script, or we have to do some manipulation like make 2 copies of data_dst pics (not aligned ones just base extract from the video) each in one folder, then paste the debug frames from the script in one folder, delete, then copy what's left into second folder, delete, and then you should have the real frames we want to correct in MVE, and not the debug frames with the red and green squares visible. Still didn't try it, but I think it should work if picture names are correct.

 

[siao]

"SPT"  "angeloshredder" "siao"
Now that i have been able to do proper tests, the script should work fine without problems (even for both JPG / PNG image formats). I edited my previous post, please refer to that.

Hello
The script is very good now
Thank you
Only one question
In the FRAMES folder
It should be the corresponding data_dst picture
But now it is the deleted aligned_debug picture
Did I understand it wrong?

---

"SPT"  "angeloshredder" "siao"
Now that i have been able to do proper tests, the script should work fine without problems (even for both JPG / PNG image formats). I edited my previous post, please refer to that.

So, I still didn't have time to try it yet, but just to be clear, the workflow should be :

Extract dst, sort & delete wrong frames, recover original file name, removing "_0" from the names, make a copy of aligned_debug folder, paste in it everything from dst_aligned, delete these pictures, now delete the pictures with no faces to re-extract, copy what's left to the real aligned_debug, delete, then use MANUAL RE-EXTRACT.

Then use your script, and it will create a folder with original full frames from the video corresponding to the frames Manual Re extract didn't manage to create. Correct ?

Hello
My understanding is
All processes must be completed in DFL first
Then "faceset MANUAL RE-EXTRACT DELETED ALIGNED_DEBUG"
Move to MVE correction

You can refer to the previous instructions
//deep.whitecatchel.ru/literotica/forums/thre...s-and-landmarks-re-extract?pid=30890#pid30890

---

If you get the pics from debug either there's still something wrong with the script, or we have to do some manipulation like make 2 copies of data_dst pics (not aligned ones just base extract from the video) each in one folder, then paste the debug frames from the script in one folder, delete, then copy what's left into second folder, delete, and then you should have the real frames we want to correct in MVE, and not the debug frames with the red and green squares visible. Still didn't try it, but I think it should work if picture names are correct.

---

I test again

It works fine when the data_dst picture is "jpg"

 

[Roy_B]

Unfortunately the problem when using PNG format still remains (i haven't got the time to test it before sorry). At the moment it's still not an easy problem to solve (for my basic programming knowledge), as it "conflicts" with the "XCOPY" parameters of the script.

If you want to use the PNG format for frame extraction, the quickest solution i know atm is to do the following steps:

1. follow Tuts mehod described in his DFL 2.0 tutorial (section "5.2 Data_dst cleanup")
2. rename the file extensions from .png to .jpg by using a dedicated software (ReNamer, Flash Renamer etc.)
3. manually delete all bad aligned_debug images to recycle bin
4. use my script
5. restore the .png extension in data_dst (after using the script)
NOTE: if you use JPG format for video extraction, you can obviously skip steps 2 and 5.



As for the workflow, i personally follow the guide created by Tuts because it allows you to have always 100% of the images (both those extracted from the video with their relative "_debug" version and all the corresponding aligned faces):
for convenience it is possible to do the same thing directly from my GUI, using the functions "PROCESS DESTINATION FACESET" and then "CLEANUP DESTINATION FACESET" buttons (always before using this script) by following the instructions contained into their tooltip buttons.
Then do the steps as described above and everything is ok (at least on my PC).
I haven't had time to test MVE's features yet but as soon as i get the chance to test them, i'll try to figure it out if there are further problems (and if I'm able to fix them through the script or in any other way).



Regarding the missing extracted faces when working with SOURCE i can't help much because it is an intrinsic problem of DFL itself (you've got to do the work manually).

 

[kakks]

Why does it keep saying "no faces found"? I have executed "5) data_dst faceset extract". and I got "aligned" and "aligned_debug" folder. and then, I put alined,aligned_debug, frame folder into MVE just like tutorial says but it showing no faces found error.

Edit: I installed all plug-ins and it worked

 

[siao]

Why does it keep saying "no faces found"?  I have executed "5) data_dst faceset extract". and I got "aligned" and  "aligned_debug" folder. and then, I put alined,aligned_debug, frame folder into MVE just like tutorial says but it showing no faces found error.  

Edit: I installed all plug-ins and it worked

I'm not sure what your problem is
I made a short video and posted it in the original topic
Please refer to that

//deep.whitecatchel.ru/literotica/forums/thread-mve-machine-video-editor-guide?pid=31022#pid31022

 

[sbep]

"no faces found" for me too. also noticed it can't detect my GPU in settings, but not sure that matters in this case. i tried installing all the plugins (damn redundant, btw... each one installs the same few gb of packages from scratch rather than co-using..), but no luck.

 

[siao]

"no faces found" for me too. also noticed it can't detect my GPU in settings, but not sure that matters in this case. i tried installing all the plugins (damn redundant, btw... each one installs the same few gb of packages from scratch rather than co-using..), but no luck.

Did you run this step?

Try it