In the Wikipedia article on eidetic memory, it's suggested that Sherlock from the latest BBC series has it, but this is then disputed with the observation that he also seems to have cultivated the ability to 'delete' things from his internal storage.

What Sherlock seems to have is pretty close to what I call an eidetic memory, i.e., whatever I've got. The sticking point seems to be whether one needs to "know" things or merely "remember" them for it to count.

There is a distinct difference between the two for me. I obviously know a lot of stuff -- this can be attested to by any one of the probably thousands of people who, over the course of my lifetime, have stared at me slack-jawed while I ramble blithely across three different subjects that normally require a graduate degree -- but that's not my eidetic memory. That's the fact that my brain is a enormous filing cabinet. Facts and procedures require understanding, at least for me. I found it completely impossible to remember anything in my college Arabic class, because the instructor would not write out anything that contained letters we hadn't learned yet, nor would she give us any grammar. It was a bunch of noise with no underpinning. These things are stored explicitly in my brain, as readable data of some kind, with no particular tie to the method of reproduction (text vs audio, print vs handwriting, male voice vs female voice) attached.

The eidetic memory is more like the thing Sherlock was doing in "A Study In Pink", when he and John were about to take off after the cab. The video overlay is a map of the immediate area, and beyond that you see flashes of street signs, landmarks, hazards, etc. These are stored as pictures and not as explicit information. They're like snapshots. A .gif of a dog doesn't have all the elements of concrete and theoretical dogness, nor is it a generalized example of an object in the state of dog. You can't look at the stream of raw data and see a dog in it; you have to bring the picture up in some kind of software that has the ability to render the data into an arrangement of colored dogs. Then you can look at it and consciously go: DOG.

Another example: A friend and I were on one of those horribly boring long drives you have to endure so often in the desert Southwest, probably either Flagstaff to Phoenix, or god forbid, Flagstaff to Las Vegas. She asked me at one point how far it was to the nearest rest stop. We had passed a sign a little while back which gave the answer she wanted, but I didn't actually remember the number -- I had to think back, pop up the mental picture and read it off of the sign. It's the difference between having a document scanned to PDF or having it retyped in a word processor. One is a bunch of pixels in an arbitrary arrangement that happens to look like letters, the other is itself a stream of manipulable information with some formatting slapped on it. My eidetic memory has no OCR at all, which is why I can't generally use it to read text straight off book pages -- although if I have already more or less memorized the text as text, I can use my memory of how the paragraph blocks were shaped and arranged to sort of follow along.

If this is how Sherlock works, then it's entirely possible for him to not really "know" how the solar system works, at the same time as being able to retrieve a diagram of same for consultation if it became important. In some cases, these snapshots are more useful than having the information explicitly available, largely because if you do things like run around London solving mysterious crimes, you often don't have any explicit information to tie to it. Something particularly suggestive of this happens in "The Blind Banker", where Sherlock starts to go on a frantic monologue about most people having blah blah craptastic visual memory before John shuts him up by pulling out his cell phone. Sherlock at some point twigged to the fact that most people do not have reference plates in their brain -- this takes a shockingly long time when this is how you've remembered things your entire life -- and is much relieved when John knows this also, and was bright enough to take a damn picture with his phone camera before it vanished.

The ability to note and recall shapes, marks, symbols, details, and movements that had no meaning at the time the memory was recorded is unbelievably handy, and I have no idea at all how other people get along without it. It's not perfect, but I cannot even begin to count how many random household items I've found by pulling up the most recent snapshot of the thing and checking the surroundings to see where I put the missing doodad down last. It also works with things like path shapes on maps, although this can get annoying because just the path shape as viewed from above doesn't give me distance or scale, nor does it automatically tie in with with the landmarks look like from street level, which means I can still get really fucking lost in places like downtown Boston.

Curiously, in order to recall visual things, I have to have my eyes open -- closing them seem to interfere with whatever circuits light up when I 'see' things that aren't right in front of me. I've no idea why. Paying too much attention to any one thing in view, ironically, can destroy the rest of the frame around it; I need to be paying attention to the world but not concentrating to get the full field-of-view snapshot. The thing that prompts the recall also has some influence over how the picture comes up. The more specific the question the more likely I am to do what everyone else does and confabulate a reasonable answer if I don't come up with one automatically. I've learned to compensate for that by bringing up the picture/video and then scanning for things so I don't burn out all the stuff I don't need, although I can still shoot myself in the foot by trying too hard.

I have it to a more or less equal degree with music, to the point where if I'm singing something a capella I have a hard time "fast-forwarding" through instrumental bridges in the backing music, to get to the part where vocals come back in, and in some cases with kinesthetics or proprioceptive feedback. One of my friends once suggested that I had perfect pitch, since as far as I can tell I sing mostly through "remembering" the positioning and degree of muscle tension that makes the noise I'm making match the memory of the noise in my head. Similar, lesser, but probably still weirdly vivid, things happen with other senses or with less-patterned sound. In the cases of smell and taste, I have less random crap in the filing cabinets, and it will sometimes drive me batty when I can taste a flavor in something or catch a scent seems really familiar, but can't immediately place it with the name or the original source.

Prodigious recall of fact is what Sherlock is doing in "The Hounds of Baskerville" when he's running through his 'memory palace' for anything and everything connected to 'liberty'. Those are explicit associations linking data to other data, not raw scans of stuff he's reading off of. The only way for those to be cross-linked like they are is for them to be encoded as explicit information with pointers to other explicit information.