about v360gopro filter 'seam'_continue here.

[suicaoya123-hub]

https://drive.google.com/drive/folders/1hR1gaY_f6sAqxoFZOneYto1iVJMYBeoq?usp=drive_link

Download the original video GS011206.360 and export it using the following command. You will then see a very obvious seam in the converted video (circled in red).

If you receive the original video, please reply to me. I will delete my video later (because it contains my personal image).

./ffmpeg -hide_banner -i GS011206.360 -filter_complex "
[0:0]format=pix_fmts=yuv420p[vf],[0:4]format=pix_fmts=yuv420p[vr],
[vf][vr]v360gopro=output=dfisheye:overlap=128:w=7680:h=3840
" -map 0:1 -map 0:3 -c:v libx265 -b:v 1500000k -c:a copy -c:u copy GS011206-dfisheye2.mp4

[aimoff]

@suicaoya123-hub
I've downloaded it.

[suicaoya123-hub]

@suicaoya123-hub I've downloaded it.

ok Hope everything goes well~

[aimoff]

It seems it is difficult to treat with single overlapped pixel size.

Original stream [0:0] is:

Magnified images are as bellow:

Overlapped area of upper-left (distance: approx. 78)

Overlapped area of lower-left (distance: approx. 86)

Overlapped area of upper-right (distance: approx. 116)

Overlapped area of lower-right (distance: approx. 98)

It may be caused by lens aberration. I can't expect perfect corection.
I'll try:
Left: rescale narrower after blending 80 (96 - 16) pixels
Right: rescale wider after blending 112 (96 + 16) pixels

[suicaoya123-hub]

It seems it is difficult to treat with single overlapped pixel size.似乎很难用单一重叠像素尺寸进行处理。

Original stream [0:0] is:原始流[0:0]为：

Magnified images are as bellow:放大后的图像如下所示：

Overlapped area of upper-left (distance: approx. 78)左上角重叠区域（距离：约 78）

Overlapped area of lower-left (distance: approx. 86)左下角重叠区域（距离：约 86）

Overlapped area of upper-right (distance: approx. 116)右上角重叠区域（距离：约 116）

Overlapped area of lower-right (distance: approx. 98)右下角重叠区域（距离：约 98）

It may be caused by lens aberration. I can't expect perfect corection.这可能是镜头像差造成的。我不能指望完全校正。 I'll try:  我会尽力： Left: rescale narrower after blending 80 (96 - 16) pixels左图：混合 80 (96 - 16) 像素后重新缩放至更窄的尺寸 Right: rescale wider after blending 112 (96 + 16) pixels右图：混合 112 (96 + 16) 像素后重新缩放至更宽

gmat/goproMax-ffmpeg-v5#5 (comment)

Perhaps you could take a look at the original code. Maybe that will give you some ideas.

(I think I encountered the same problem as him.)

Perhaps the gap is caused by an incorrect projection approach.

Here's my thought: (It might not be correct)

As shown in red, assuming the original fisheye is 190 degrees, but after EAC , its 190-degree position should be the blue line position of the lens.

Normally, the overlap we crop is the yellow line. But the actual landing point should be on the red line (the red line might not be that exaggerated in reality), because the 190-degree edge of the fisheye is actually straightened to the blue line. Therefore, the fov180 part of the red line is the real cropping position.

So, based on this, we should restore the sphere projection, and then crop the excess fov, which might allow for alignment.

Of course, my idea might be wrong; the yellow line is indeed a straight line, which is the fov180 position.

[aimoff]

I'll try:
Left: rescale narrower after blending 80 (96 - 16) pixels
Right: rescale wider after blending 112 (96 + 16) pixels

I pushed tentative code to gopro/8.0-max2 branch.

Exported EAC image (top only) is as bellow:

$ ./ffmpeg -hide_banner -i GS011206.360 -filter_complex "
[0:0][0:4]v360gopro=output=eac:overlap=96:w=5760:h=3840
" -map 0:1 -map 0:3 -c:v libx265 -b:v 1500000k -c:a copy -c:u copy GS011206-eac.mp4

It seems almost good, but not perfect.

[suicaoya123-hub]

I'll try:  我会尽力：
Left: rescale narrower after blending 80 (96 - 16) pixels左图：混合 80 (96 - 16) 像素后重新缩放至更窄的尺寸
Right: rescale wider after blending 112 (96 + 16) pixels右图：混合 112 (96 + 16) 像素后重新缩放至更宽

I pushed tentative code to gopro/8.0-max2 branch.我已将初步代码推送到 gopro/8.0-max2 分支。

Exported EAC image (top only) is as bellow:导出的 EAC 图像（仅顶部）如下所示：

$ ./ffmpeg -hide_banner -i GS011206.360 -filter_complex "
[0:0][0:4]v360gopro=output=eac:overlap=96:w=5760:h=3840
" -map 0:1 -map 0:3 -c:v libx265 -b:v 1500000k -c:a copy -c:u copy GS011206-eac.mp4
It seems almost good, but not perfect.它看起来几乎完美，但还不够完美。

That's great!

However, according to the official documentation for Maxsun 2, the ideal resolution for stereo images exported from EAC is 7680x3840 (2:1).

This is because twice the height of the EAC (top part) is the optimal resolution for a true fisheye view. This allows for better visualization of pixel-to-pixel details and facilitates the inspection of seam issues.

[aimoff]

However, according to the official documentation for Maxsun 2, the ideal resolution for stereo images exported from EAC is 7680x3840 (2:1).

No. EAC's aspect ratio is 3:2. 6 cubes are lined up in 2 rows.
For dfisheye, 7680x3840 (2:1) is correct.

[suicaoya123-hub]

However, according to the official documentation for Maxsun 2, the ideal resolution for stereo images exported from EAC is 7680x3840 (2:1).

No. EAC's aspect ratio is 3:2. 6 cubes are lined up in 2 rows. For dfisheye, 7680x3840 (2:1) is correct.

seams still obviously .this is hard🤣···

[aimoff]

@suicaoya123-hub
Probably, we need another logic like:

As shown in red, assuming the original fisheye is 190 degrees, but after EAC , its 190-degree position should be the blue line position of the lens.

It is difficult to find it in short term.

I plan to submit the code with minimal changes to support 10bit color depth to FFmpeg community.

[suicaoya123-hub]

@suicaoya123-hub Probably, we need another logic like:

As shown in red, assuming the original fisheye is 190 degrees, but after EAC , its 190-degree position should be the blue line position of the lens.

It is difficult to find it in short term.

I plan to submit the code with minimal changes to support 10bit color depth to FFmpeg community.

Here at

gmat/goproMax-ffmpeg-v5#5 (comment)

someone mentioned that this library

https://github.com/trek-view/max2sphere outputs seamless . I think you could refer to its code logic. If it works...