search

nagadomi / nunif

Misc; latest version of waifu2x; 2D video to stereo 3D video conversion
MIT License
1.42k stars 134 forks source link

my 3d is backwards in the words of the clips #197

Closed andy500 closed 1 month ago

andy500 commented 2 months ago

how can your program create the words in the video clip back to front in every frame can your program reprogram itself and don't think l say it likely ?

nagadomi commented 2 months ago

Front and back cannot be swapped. Because the real world doesn't see that. If iw3 results look like that, then there is a bug in your video player/device. The right and left eyes are reversed.

andy500 commented 2 months ago

it is not the video player/device every word in this video is in reverse not the people this is the link https://drive.google.com/file/d/1HSrlO4GGzgeLtjUtDjWknxaQg-mcJAHN/view?usp=drive_link

On Mon, 29 Jul 2024 at 23:16, nagadomi @.***> wrote:

Front and back cannot be swapped. Because the real world doesn't see that. If iw3 results look like that, then there is a bug in your video player/device. The right and left eyes are reversed.

— Reply to this email directly, view it on GitHub https://github.com/nagadomi/nunif/issues/197#issuecomment-2257103305, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFIOA3JVIVCMORIZJYSYYOTZO25M7AVCNFSM6AAAAABLVC3DZGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDENJXGEYDGMZQGU . You are receiving this because you authored the thread.Message ID: @.***>

nagadomi commented 2 months ago

Please provide examples in public. I will not check that link.

andy500 commented 2 months ago

every word in this video is in reverse not the people and it is not a video player/device issues and l can not create it again your program as create a one of video that is in possible to create in 30min which is what your program as create you would need to see the video to understand it

nagadomi commented 2 months ago

I never request permission to access the file. because of security reasons.

Does WORD mean subtitle? If so, that is depth model issue and is not currently supported.

andy500 commented 2 months ago

no it is not subtitle issues it is just thet every frame with a word in it is in reverse every street name in the move is in reverse every word what ever it is on is in reverse l never seen a video like it

nagadomi commented 2 months ago

Do you mean that all frames are horizontally flipped? And the original video is not. hflip

andy500 commented 2 months ago

sorry no just every words on every thing in the movie is in reverse nothing else l think you just have to look at the clip

nagadomi commented 2 months ago

Could you post a screenshot?

diverswan9 commented 2 months ago

@andy500 in your google-drive for the file, under 'share' / 'share' - change the 'General Access' from 'Restricted' to 'Anyone with the link'. It will be easier for people to view the file.

andy500 commented 2 months ago

ok and thank you for your new update much faster to convert to 3d one question do l need anything v-2_L model Python code all can l use the python code in the pretrained_models hub

diverswan9 commented 2 months ago

2024-07-30_093406

His 'source file' is flipped so everything is like in a mirror but the 3d is correct. So @andy500 wants to 'flip' the source file before the 3d processing happens. Maybe a filter setting?

@nagadomi , which reminds me... do the filter settings in '-vf (src) happen before or after the 3d processing?

diverswan9 commented 2 months ago

@andy500 enter hflip into filter settings in '-vf (src). This will horizontally 'flip' the source video.

2024-07-30_094921

nagadomi commented 2 months ago

@diverswan9 --vf is an option for the input video stream. It is applied before depth estimation. hflip should resolve this issue as you say.

diverswan9 commented 2 months ago

@diverswan9 --vf is an option for the input video stream. It is applied before depth estimation. hflip should resolve this issue as you say.

Yeah, I just found that out by testing hflip. Thanks!

diverswan9 commented 2 months ago

one question do l need anything v-2_L model Python code all can l use the python code in the pretrained_models hub

@andy500 - The Depth_Anything_V2 models need to be manually downloaded and put in 'Checkpoints' folder.

     \nunif-windows\nunif\iw3\pretrained_models\hub\checkpoints

https://huggingface.co/depth-anything

andy500 commented 2 months ago

l put The Depth_Anything_V2 models in the Checkpoints folder is that all l need to do no l did not use vf setting that how it come out

diverswan9 commented 2 months ago

l put The Depth_Anything_V2 models in the Checkpoints folder is that all l need to do no l did not use vf setting that how it come out

@andy500 - The V2 models should show up now in 'Depth Model'.

To get your video to 'display' the correct way enter hflip in '-vf' box before processing. For some reason your 'source' video is horizontally flipped.

andy500 commented 2 months ago

no more problems just did another clip come out ok thanks l had to add the V2 models in the gui did not show in Depth Model after restart the program but does now

andy500 commented 1 month ago

Awesome, thanks for the update nagadomi

On Fri, 2 Aug 2024 at 09:08, nagadomi @.***> wrote:

Closed #197 https://github.com/nagadomi/nunif/issues/197 as completed.

— Reply to this email directly, view it on GitHub https://github.com/nagadomi/nunif/issues/197#event-13738066223, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFIOA3LNLAQHNSEMLLRC2V3ZPM5BRAVCNFSM6AAAAABLVC3DZGVHI2DSMVQWIX3LMV45UABCJFZXG5LFIV3GK3TUJZXXI2LGNFRWC5DJN5XDWMJTG4ZTQMBWGYZDEMY . You are receiving this because you were mentioned.Message ID: @.***>