Difference between revisions of "Talk:2414: Solar System Compression Artifacts"

Explain xkcd: It's 'cause you're dumb.
Jump to: navigation, search
(explaining dark color banding)
(edit to previous comment)
Line 2: Line 2:
 
"([Compression artefacts] may become literally unnoticeable because hexadecimal color values are discrete[...]" - disagree. Artefacts exist ''because'' of a discrete nature. Either of the RGB(/HSV/whatever) granularity, the lower the colour depth, or of the method used to get around the overheads of storing literal 24+ bits of colour-depth across a given image size. TrueColo(u)r should escape ''perceived'' colour-banding, but any image editor knows (or relies upon) that any flood-fill/by-colour-selection used with an absolute drift range away from the datum spot less than that across a gradient spills away from it can highlight 'hidden' edges between (say) <span style="color:#789ABC">#789ABC</span> and <span style="color:#789BBC">#789BBC</span>. What we have here is low bit-depth (grey-shade or equal-RGB, apparently 4-bit(/each), counting the 16 bands thanks to the mach-banding) non-dithered band-shading of a possibly nuanced (fractal?) shade fall-off. Possibly a 2D slice through 3D (or more, e.g. if animated) of voxelated (or hypervoxelated) stored values, which use up a ''lot'' of space in the Universe Simulator. Perhaps there's also something like Discrete Cosine Transform compression for easier block/chunk storage, retrieval and/or generation-on-demand (with detailed deltas for complex overlaying features such as Voyager). Because the Creator/Programmer of the universe has limited storage/processor cycles! [[Special:Contributions/141.101.105.122|141.101.105.122]] 01:39, 21 January 2021 (UTC)
 
"([Compression artefacts] may become literally unnoticeable because hexadecimal color values are discrete[...]" - disagree. Artefacts exist ''because'' of a discrete nature. Either of the RGB(/HSV/whatever) granularity, the lower the colour depth, or of the method used to get around the overheads of storing literal 24+ bits of colour-depth across a given image size. TrueColo(u)r should escape ''perceived'' colour-banding, but any image editor knows (or relies upon) that any flood-fill/by-colour-selection used with an absolute drift range away from the datum spot less than that across a gradient spills away from it can highlight 'hidden' edges between (say) <span style="color:#789ABC">#789ABC</span> and <span style="color:#789BBC">#789BBC</span>. What we have here is low bit-depth (grey-shade or equal-RGB, apparently 4-bit(/each), counting the 16 bands thanks to the mach-banding) non-dithered band-shading of a possibly nuanced (fractal?) shade fall-off. Possibly a 2D slice through 3D (or more, e.g. if animated) of voxelated (or hypervoxelated) stored values, which use up a ''lot'' of space in the Universe Simulator. Perhaps there's also something like Discrete Cosine Transform compression for easier block/chunk storage, retrieval and/or generation-on-demand (with detailed deltas for complex overlaying features such as Voyager). Because the Creator/Programmer of the universe has limited storage/processor cycles! [[Special:Contributions/141.101.105.122|141.101.105.122]] 01:39, 21 January 2021 (UTC)
 
:tl;dr?<span> β€” [[User:Sqrt-1|The <b>π—¦π—Ύπ—Ώπ˜-𝟭</b>]] <sup>[[User talk:Sqrt-1|<span style="color: blue">talk</span>]] [[Special:Contributions/Sqrt-1|<span style="color: blue">stalk</span>]]</sup></span> 13:16, 21 January 2021 (UTC)
 
:tl;dr?<span> β€” [[User:Sqrt-1|The <b>π—¦π—Ύπ—Ώπ˜-𝟭</b>]] <sup>[[User talk:Sqrt-1|<span style="color: blue">talk</span>]] [[Special:Contributions/Sqrt-1|<span style="color: blue">stalk</span>]]</sup></span> 13:16, 21 January 2021 (UTC)
βˆ’
::TLDR is this Tom Scott video on the topic: https://youtu.be/h9j89L8eQQk. Short version: the difference between blacks #010101 and #020202 is more noticeable than the difference between whites #FEFEFE and #FDFDFD. If your picture is dark, and especially if it is compressed, you will often get ugly bands of different shades of black. [[Special:Contributions/108.162.237.28|108.162.237.28]] 16:50, 22 January 2021 (UTC)
+
::TLDR is this Tom Scott video on the topic: https://youtu.be/h9j89L8eQQk. Short version: the difference between blacks #010101 and #020202 (a doubling of brightness) is more noticeable than the difference between whites #FEFEFE and #FDFDFD (a 0.00001% change in brightness). If your picture is dark, and especially if it is compressed, you will often get ugly bands of different shades of black. [[Special:Contributions/108.162.237.28|108.162.237.28]] 16:50, 22 January 2021 (UTC)
 
I would suggest that more emphasis needs to be placed on 'dynamic range' and 'undetectable' in this explanation. Particularly noticeable in streaming video codecs, you often can't decipher any information in dark scenes/regions. So the joke is that the map beyond here is empty, mostly because it is too far down in the dynamic range of our lossy observations. [[Special:Contributions/108.162.219.80|108.162.219.80]] 17:36, 21 January 2021 (UTC)
 
I would suggest that more emphasis needs to be placed on 'dynamic range' and 'undetectable' in this explanation. Particularly noticeable in streaming video codecs, you often can't decipher any information in dark scenes/regions. So the joke is that the map beyond here is empty, mostly because it is too far down in the dynamic range of our lossy observations. [[Special:Contributions/108.162.219.80|108.162.219.80]] 17:36, 21 January 2021 (UTC)
 
:Re "...stretches out over maybe a dozen such low-res pixels/AUs, which is equivalent to slightly more than the radius of Saturn's orbit or the entire diameter of Jupiter's!", this assumes that Jupiter and Voyager are the same distance from the imaginary "camera". I can completely cover the moon with my thumb, but that does not imply that they are similar in size, because my thumb is closer to my eye. ''(Unsigned!)''
 
:Re "...stretches out over maybe a dozen such low-res pixels/AUs, which is equivalent to slightly more than the radius of Saturn's orbit or the entire diameter of Jupiter's!", this assumes that Jupiter and Voyager are the same distance from the imaginary "camera". I can completely cover the moon with my thumb, but that does not imply that they are similar in size, because my thumb is closer to my eye. ''(Unsigned!)''

Revision as of 16:52, 22 January 2021

"([Compression artefacts] may become literally unnoticeable because hexadecimal color values are discrete[...]" - disagree. Artefacts exist because of a discrete nature. Either of the RGB(/HSV/whatever) granularity, the lower the colour depth, or of the method used to get around the overheads of storing literal 24+ bits of colour-depth across a given image size. TrueColo(u)r should escape perceived colour-banding, but any image editor knows (or relies upon) that any flood-fill/by-colour-selection used with an absolute drift range away from the datum spot less than that across a gradient spills away from it can highlight 'hidden' edges between (say) #789ABC and #789BBC. What we have here is low bit-depth (grey-shade or equal-RGB, apparently 4-bit(/each), counting the 16 bands thanks to the mach-banding) non-dithered band-shading of a possibly nuanced (fractal?) shade fall-off. Possibly a 2D slice through 3D (or more, e.g. if animated) of voxelated (or hypervoxelated) stored values, which use up a lot of space in the Universe Simulator. Perhaps there's also something like Discrete Cosine Transform compression for easier block/chunk storage, retrieval and/or generation-on-demand (with detailed deltas for complex overlaying features such as Voyager). Because the Creator/Programmer of the universe has limited storage/processor cycles! 141.101.105.122 01:39, 21 January 2021 (UTC)

tl;dr? β€” The π—¦π—Ύπ—Ώπ˜-𝟭 talk stalk 13:16, 21 January 2021 (UTC)
TLDR is this Tom Scott video on the topic: https://youtu.be/h9j89L8eQQk. Short version: the difference between blacks #010101 and #020202 (a doubling of brightness) is more noticeable than the difference between whites #FEFEFE and #FDFDFD (a 0.00001% change in brightness). If your picture is dark, and especially if it is compressed, you will often get ugly bands of different shades of black. 108.162.237.28 16:50, 22 January 2021 (UTC)

I would suggest that more emphasis needs to be placed on 'dynamic range' and 'undetectable' in this explanation. Particularly noticeable in streaming video codecs, you often can't decipher any information in dark scenes/regions. So the joke is that the map beyond here is empty, mostly because it is too far down in the dynamic range of our lossy observations. 108.162.219.80 17:36, 21 January 2021 (UTC)

Re "...stretches out over maybe a dozen such low-res pixels/AUs, which is equivalent to slightly more than the radius of Saturn's orbit or the entire diameter of Jupiter's!", this assumes that Jupiter and Voyager are the same distance from the imaginary "camera". I can completely cover the moon with my thumb, but that does not imply that they are similar in size, because my thumb is closer to my eye. (Unsigned!)
I presume that this is the 'data' version of our system, straight from the 'cosmic computer' behind reality, viewed orthographically like a Minecraft map. But it matters not. The text quoted clearly gives scale context (on the justifiable presumption that the low-res grid is a 1AU-sized display of solar particle distribution) that "the range of the map the Voyager covers is like the size of these orbits". Just like "the area of the sky my thumb obscures is roughly a Moon-width". 141.101.104.241 01:18, 22 January 2021 (UTC)
Re: "The Voyager image (and track) is overlaid at finer resolution", if it not unheard-of for a compression algorithm to render, say, 32x32 pixel blocks as if they were single, larger pixels if their immediate neighbors are almost the same shade, while rendering small, detailed, and high contrast portions of the image at a higher resolution. No "overlay" required. (Still unsigned!)
Within an area of finer details, artefacts would be seen as the "meh, this is just one block" attitude is changed to incorperate detail of interest within a sub-block. There's no sign of fringe-artefacts (other than normal XKCS antialiasing of lines against background). I'd say it was a "solar wind" low-detail layer over which is incorporated a "Voyager" standard-detail layer with transparency, not a single layer of subject-aware tunable blockwise compression. (It might have come out differently if the composition was saved in an actual lossy-compression, by Randall, rather than .png, but for different reasons. Bit that's a meta-discussion issue, not nerd-sniping.) 141.101.104.241 01:18, 22 January 2021 (UTC)

This doesn't just happen in old images. I still see this frequently when watching movies, even new ones, where what should be a smooth gradation of tones shows steps. The other obvious defect is poor sprite handling, which causes artifacts like someone's facial features not moving with their head movements. I'm not sure dithering would be a useful way to dispel the banding, since dithering would increase the compressed data size because you no longer have large areas of all-the-same-pixel-value. The banding could also be an artifact of the decompression. If you consider that a format like JPEG uses something like Fourier transforms it seems it should be able to represent a gradation easily and the stepped banding with difficulty, so I might be tempted to blame the banding on the decompression code.108.162.241.70 13:46, 22 January 2021 (UTC)

Source of Voyager 1

Given Voyager 1's distance and age now, wouldn't it be more appropriate to change the opening explanatory sentence to "Voyager 1 is a space probe launched from Earth in 1977." The only sentient creatures who may still care about its origins in the United States are probably also from the United States. The next sentient beings to encounter Voyager 1 will have no understanding of what the United States is or was.