EVA-EXTRA Free Paper Magazine
Moderators: Rebuild/OT Moderators, Board Staff
Forum rules
By visiting this forum, you agree to read the rules for discussion.
By visiting this forum, you agree to read the rules for discussion.
- Josh_LawLiet
- Embryo
- Age: 33
- Posts: 13
- Joined: Jul 13, 2012
- Location: Tokio-3
- Gender: Male
Hello It's me again, I want to let you know that I' m still working on the clean up. Thanks for cleaning the audio tomfrom994, one thing less to do.
I'm stuck at clean up black lines while I look at 2 valid choices:
1- Remove black lines and reunite the image (optional use JPEG smoothing algorythm for better result). Status: Done
2- Instead of removing black lines, fill them with interpolation of other pixels near the black zone. Status: In Process
I will upload soon both results and let you decide which improve better the image.
If someone have thought about other ways or wanna try by himself you can use the image 6 post up (the one with perspective view corrected).
When we got the image cleaned right, MATLAB will use appropiate algorythm clean automatically (Hardest step) the 3900 images of the entire video then we will use the cleaned audio and the cleaned images in order to mount the video again (Easy step).
I'm stuck at clean up black lines while I look at 2 valid choices:
1- Remove black lines and reunite the image (optional use JPEG smoothing algorythm for better result). Status: Done
2- Instead of removing black lines, fill them with interpolation of other pixels near the black zone. Status: In Process
I will upload soon both results and let you decide which improve better the image.
If someone have thought about other ways or wanna try by himself you can use the image 6 post up (the one with perspective view corrected).
When we got the image cleaned right, MATLAB will use appropiate algorythm clean automatically (Hardest step) the 3900 images of the entire video then we will use the cleaned audio and the cleaned images in order to mount the video again (Easy step).
Where ever Asuka (Soryu) is, I will manage to get there.
- tomfrom994
- Adam
- Age: 29
- Posts: 69
- Joined: Jun 02, 2011
- Gender: Male
- Josh_LawLiet
- Embryo
- Age: 33
- Posts: 13
- Joined: Jul 13, 2012
- Location: Tokio-3
- Gender: Male
As I promised here you have the 2 ways of cleaning, please judge it and tell me how you want the video with remove method or interpolation one.
Personally I prefer way of interpolation but if you all decide the remove way I will make the video with it.
I prefer 2nd method, as you can see still some black lines are in it, this is because the mask I used for interpolate pixels isn't perfect, when it is, there will be no black lines.
We won't obtain perfect image because we cant recreate the original pixels that weren't on it.
When you decide which is better MATLAB will automatize the process for all 3906 captions of the video ( yeh I love Asuka, evangelion and so but I WONT do it manually XD)
Personally I prefer way of interpolation but if you all decide the remove way I will make the video with it.
Image examples SPOILER: Show
Original Caption of the video "Goddess Asuka":
Perspective Restored with ratio method improved (1920x1080 like the original):
Now the 2 ways of cleaning black lines.
1- REMOVE BLACK LINES METHOD (as you can see ratio is lost):
2- INTERPOLATE BLACK LINES METHOD (ratio retained, pixels aproximated by near zones values)
I prefer 2nd method, as you can see still some black lines are in it, this is because the mask I used for interpolate pixels isn't perfect, when it is, there will be no black lines.
We won't obtain perfect image because we cant recreate the original pixels that weren't on it.
When you decide which is better MATLAB will automatize the process for all 3906 captions of the video ( yeh I love Asuka, evangelion and so but I WONT do it manually XD)
Where ever Asuka (Soryu) is, I will manage to get there.
- Sailor Star Dust
- Kept you waiting, huh?
- Age: 38
- Posts: 23063
- Joined: Aug 13, 2006
- Location: 私の中いる自分の心
- Gender: Female
- Blue Monday
- Angel
- Age: 34
- Posts: 3363
- Joined: Jun 17, 2012
- Location: Earth-33
- riffraff11235
- Seed of Life
- Age: 31
- Posts: 3975
- Joined: Jul 19, 2012
- Location: Long Island, New York, USA
- Gender: Male
Yes! The second method looks great! Once you get the video running, I'm sure the interpolated pixels will be barely noticeable. Keep it up!
だから みんな 死んでしまえば いいのに... では, あなたは何故, ココにいるの? ...ココにいても, いいの?
"Cogito ergo sum." - Rene Descartes
"Men do not quit playing because they grow old; they grow old because they quit playing." - Oliver Wendell Holmes
Avatar: Asuka is superior. That is all.
PSN - riffraff-11235 Feel free to add me. PM me on EGF if you do.
Steam - rifffraff11235
League of Legends - riffraff11235
Osu! - riffraff11235
"Cogito ergo sum." - Rene Descartes
"Men do not quit playing because they grow old; they grow old because they quit playing." - Oliver Wendell Holmes
Avatar: Asuka is superior. That is all.
PSN - riffraff-11235 Feel free to add me. PM me on EGF if you do.
Steam - rifffraff11235
League of Legends - riffraff11235
Osu! - riffraff11235
- TMBounty_Hunter
- Hachigouki is my waifu
- Posts: 3230
- Joined: Nov 08, 2008
- Location: Toronto, Canada
- Gendo'sPapa
- Committeeperson
- Age: 38
- Posts: 5599
- Joined: Oct 24, 2006
- Gender: Male
- Warren Peace
- Potential Pilot
- Age: 93
- Posts: 1789
- Joined: May 28, 2010
- Josh_LawLiet
- Embryo
- Age: 33
- Posts: 13
- Joined: Jul 13, 2012
- Location: Tokio-3
- Gender: Male
All votes to interpolation good choice. As Gendo's Papa said I'm doing this because I consider that a challenge, and resolving It fills some feelings I have towards evangelion series, sure some of you have them also.
Now talking about the process I'm on last and hardest thing: Automatize all the process, the main problem is a big one, camera moves = image moves = black lines moves = mask must be moved in order to match black lines for all images.
Hard but not imposible
The only fear I have is that my PC get burnt as 1 image requires 100 % CPU and like 20 seconds to process it, the 3900 images will take long with 100 % CPU but maybe I can put it on a icebox
Now talking about the process I'm on last and hardest thing: Automatize all the process, the main problem is a big one, camera moves = image moves = black lines moves = mask must be moved in order to match black lines for all images.
Hard but not imposible
The only fear I have is that my PC get burnt as 1 image requires 100 % CPU and like 20 seconds to process it, the 3900 images will take long with 100 % CPU but maybe I can put it on a icebox
Where ever Asuka (Soryu) is, I will manage to get there.
- tomfrom994
- Adam
- Age: 29
- Posts: 69
- Joined: Jun 02, 2011
- Gender: Male
- Reichu
- Admin Emeritus
- Posts: 24046
- Joined: Aug 21, 2004
- Location: Sailing for the white shores
- Gender: Female
- Contact:
Eh. Even if the whole thing gets released eventually online or as a BD extra, it's a way to flex problem-solving skills and learn new things. Nothing wasteful in that.
さらば、全てのEvaGeeks。
「滅びの運命は新生の喜びでもある」
Departure Message | The Arqa Apocrypha: An Evangelion Analysis Blog
「滅びの運命は新生の喜びでもある」
Departure Message | The Arqa Apocrypha: An Evangelion Analysis Blog
- Josh_LawLiet
- Embryo
- Age: 33
- Posts: 13
- Joined: Jul 13, 2012
- Location: Tokio-3
- Gender: Male
Hello It's me, sorry for no updates but all that time I wasn't in my country and cant work on the final step also I was relaxing in nature place where you shouldn't have any digital stuff.
So now I'm on home and I will keep trying last step, It's really hard to complete because pixels moves and I need to move mask in order to match black lines every frame, and the most hard thing is that the movement is not the same, so We are in very hard problem xD
I will keep thinking how to solve that.
So now I'm on home and I will keep trying last step, It's really hard to complete because pixels moves and I need to move mask in order to match black lines every frame, and the most hard thing is that the movement is not the same, so We are in very hard problem xD
I will keep thinking how to solve that.
Where ever Asuka (Soryu) is, I will manage to get there.
- riffraff11235
- Seed of Life
- Age: 31
- Posts: 3975
- Joined: Jul 19, 2012
- Location: Long Island, New York, USA
- Gender: Male
I'm still really surprised that the original trailer hasn't been posted online yet. That said, I'm still eagerly looking forward to the fruits of your labor, Josh!
だから みんな 死んでしまえば いいのに... では, あなたは何故, ココにいるの? ...ココにいても, いいの?
"Cogito ergo sum." - Rene Descartes
"Men do not quit playing because they grow old; they grow old because they quit playing." - Oliver Wendell Holmes
Avatar: Asuka is superior. That is all.
PSN - riffraff-11235 Feel free to add me. PM me on EGF if you do.
Steam - rifffraff11235
League of Legends - riffraff11235
Osu! - riffraff11235
"Cogito ergo sum." - Rene Descartes
"Men do not quit playing because they grow old; they grow old because they quit playing." - Oliver Wendell Holmes
Avatar: Asuka is superior. That is all.
PSN - riffraff-11235 Feel free to add me. PM me on EGF if you do.
Steam - rifffraff11235
League of Legends - riffraff11235
Osu! - riffraff11235
Why are you bothering with the shaky hand held video? Why not use the steady one here:
The hand held video does have better quality though since it is not from such a skewed perspective.
If you want to go with the hand held video you're going to need to code a pixel based feature tracker across frames. This is a pain since there's not a lot of static features, too dark, lots of noise, variable brightness of projected video distorts the luminosity of surrounding building features, and so on.
If you want the hand-held one because it has better video, you can always do it like this:
1) Take the steady one posted above, and slice out the video from the rest of the building. It should be easy since the camera is not moving much. Then transform the perspective for that slice.
2) Process the shaky one, and transform the whole image including the building by a best-guess-perspective. Do this because you necessarily do not know where in the image is the projected video itself.
3) Scan for the frame gotten from the steady video in the transformed shaky one. You can brute force this by sliding the frame from step 1 over the frame in step 2, then finding the point of lowest difference.
4) Use the coordinates acquired from step 3 to rip out the larger image from the shaky video.
In and of itself, self-evident.
Return to “Rebuild of Evangelion Discussion”
Who is online
Users browsing this forum: No registered users and 16 guests