A press release from Sigma I saw on the "Imaging-Resource" site:
http://www.imaging-resource.com/NEWS/1195002045.html
 
Have only skimmed the announcement - but given the interest of late in circular fisheye lens I thought to post it here for your own investigation.
 
For us APS-size sensor users (Nikon D70/80/300 & Canon 20/30/40D, etc.) the options have apparently opened up - full circle fish eye images (f/2.8; 4.5mm focal length)!
 
Didn't seen a price in this press release.

You beat me to it Jim.
I've just seen the news item on DPReview
Here:
http://www.dpreview.com/news/0711/ [...] 45fish.asp

i've tried Sigma UK and my local stockist. Neither have any idea as to what the pricing will be.
 
Anyone care to hazard a guess?
 
JohnP

The price has not been announced yet. I'm waiting for Sigma to give me more info and I'll be updating my website as soon as I get it. I've got a unit on order for testing so will post the results here.
 
At the moment, I've no clarification on if it's going to give a full circle on a reduced sensor frame and/or what the field of view will be.

Andrew
I for one will be looking forward to hearing your comments.
 
According to the press release the lens will produce a Circular image with 180° FOV on APS-C sensors

Indeed, but no-one at Sigma seems to know this yet!

That would be a lot like the spec that the Nikkor 10.5mm has a 180° FOV from corner to corner.
 
But when I played with one on a D80 I couldn't see my hand at the corner like I can when I hold it out to the side of my Coolpix/FC-E9.
 
I'm still looking forward to what you find out.
 
I'll cross my fingers

Hi to All!
 
Have had a chance to read between the lines re: the 4.5mm - on Sigma's "Lenses" page, best accessed from the "News" tab on their site: http://sigmaphoto.com/news/news.asp
 
There also has been an announcement of a Sigma 10.0mm fisheye. In this announcement they breakout the angle of view into three categories: Nikon (180 d), Canon (167 d), & Sigma (154 d).
 
This is not done for the 4.5mm - possibly an oversight, but seems unlikely. So a possible translation would be the 4.5's image circle would easily fit on Canon (1.6x crop) & Nikon (1.5x crop) sensors.
 
Also, for the 4.5mm, Sigma mentions scientific applications such as astrophotography. Here the reference is specifically directed to the f/2.8 spec. However, it doesn't matter if it is f/1.4 if the accutance of the lens is not very good - and Sigma gives a nod in this direction with their reference to "high definition and good image quality". Perhaps Sigma is taking a page from the Nikkor 10.5mm - emphasize the sharpness and let computer software deal with distortion correction (distortion vs. sharpness, the trade-off in consumer grade fisheye lens design until the for-digital market).
 
I look forward to badders' testing results.
 
 
======================
Price: Still could find no mention...
 
Perhaps we could have a betting pool - closest to the MSRP - before Sigma pubicly/privately announces it? (Paid through PayPal.) My vote: the same as the Sigma 8mm. I'm in for US$10.

On C/Net News they mention a price for the two lenses in Japan
 
"Judging by the respective Japanese Web sites, the 10mm model will cost 90,500 yen, or about $820, and be available in December, and the 4.5mm model will cost about $1,040."
 
Its on http://www.news.com/8301-13580_3-9 [...] tag=repblg

Any idea if the new Sigma 4.5mm could work with Stitcher DS and a full frame Canon 5D ? I've just bought the older full frame 8mm Sigma EX DG F3.5 to produce 3 circular fish eye frames. Will the new one do this in two frames as it'll have over 180 degree coverage image circle ?
I know some people use the Nikon DX 10.5mm fisheye (without lens shade) to make three frame stitches on a 5D ... Others use the vastly expensive Coastal Optics 4mm lens.  
Anyone clever enough to know the answer yet ?! ;o)

Hi Harrison!
 
From what I gather the max FOV for the 4.5mm Sigma is 180 degrees, no matter if you have a full frame sensor, so Stitcher DS does not appear to be option with this lens. I base this on Sigma's web site breaking out the FOV numbers differently for Nikon, Canon, & Sigma bodies/sensors for their new 10mm fisheye, but does not do so for the 4.5mm. Assumming this is not an oversight on Sigma's part, it must mean that the 4.5mm image circle fits on a Sigma sensor (which is smaller than Nikon's APS size sensor). Yet the FOV for the 4.5mm is spec'd at 180 degrees for all camera bodies/sensors, i.e. even though the image circle is smaller than the Nikon sensor it still only produces a 180 degree image.
 
A bit of conjecture based on limited info, but if accurate, this lens doesn't appear to offer a benefit for the 5D. And since the image is full circle, giving the 4.5 a shave wouldn't seem to make a difference in its FOV.
 
But for a Nikon 10.5mm/APS user such as myself this lens looks very interesting - three shots around and probably won't need the caps for most panos with the right tilt. "For Hire" VRs with high action content become a consideration (even if a tad on the small side, possibly 3500 px spherical/875 px cube face?).

Jim:
------------
Perhaps we could have a betting pool - closest to the MSRP - before Sigma pubicly/privately announces it? (Paid through PayPal.) My vote: the same as the Sigma 8mm. I'm in for US$10.

Doug:
-----------
C/Net quote:
Judging by the respective Japanese Web sites, the 10mm model will cost 90,500 yen, or about $820, and be available in December, and the 4.5mm model will cost about $1,040."
 
Sigma's 8mm MSRP price: $990
 
Off by $50...

Thanks for your wisdom again Jim ... reckon you are right ... shame though ... would love to try some high quality full screen two frame stitches of action etc but this doesn't seem possible with the new Sigma 4.5mm lens. Guess time will tell for sure ...
Out of interest, is a shaved Nikon 10.5mm much better quality than my Sigma 8mm EX DG on the 5D with a lens adaptor ? I notice the Sigma 8mm is not as sharp as the Canon 15mm full frame fisheye so my full screen QTVR's lack resolution and detail - but are quicker to make.

Hi Harrison,
 
To clarify, the new Sigma 4.5mm will NOT allow 2 shot stitching in REALVIZ Stitcher DS. You need a lens with GREATER that 180 degs field of view to allow an overlap to blend the seam area.
 
If you want full screen high res - take more pictures.

For those not quite understanding about 180° fisheye lenses, you have to have more than 180° Field of View, usually 183° or more, to be able to stitch 2 circular fisheyes. The overlap is where the stitching software aligns and blends the 2 images.  
 
From its description the new Sigma 4.5mm produces the same 180° Circular Image on an APS-C sized sensor (approx 23x15mm) that the Sigma 8mm can produce on a Full Frame sensor (approx 36x24mm).
 
But its still just a 180° circle, so there's no overlap to align and blend.
 
Theoretically, the Sigma 4.5mm could be used in a 3-shot capture with a 10-15° tilt up eliminating the need for a Zenith shot.
 
A few years ago digital cameras fell into 3 broad catagories; (1)basic point & shoots, (2)"pro-consumer" point & shoots like the Nikon Coolpix line and the Canon G series and (3) Digital Single Lens Reflex cameras that were very expensive.
 
With lower priced dSLRs, the new trend in cameras & lenses seems to be that the "affordable dSLRs" with APS-C sensors are replacing the "pro-consumer" cameras at about the same prices. A Nikon Coolpix 8700 was $999.95 when it was new, now a Nikon D80 is about $900. And part of that trend is the lens manufacturers are making lenses specifically for the APS-C sensors like these new lenses from Sigma.
 
Andrew is right, if you want more resolution, you need to start with more resolution. That is, more pixels to stitch together.
 
Here's an explanation but the numbers are just for example
 
Think of an average spherical image for real estate as having about 5 million pixels projected onto the interior of the sphere. (with compression the file is about 300-400kb) So the sphere is about 2048 pixels around horizontal and 1024 pixel from Nadir to Zenith following the interior of the sphere. It displays well at 320 x 240 pixels
 
If you want to display a crisp, sharp interactive image at 640 x 480 you'll need at least 20 million pixels (with compression the file won't be anywhere near that size) to project onto the interion of a sphere 4096 x 2048 pixels.
 
The bigger you make the sphere (or cube) the more pixels you need to start with, before stitching.  
 
New higher resolution cameras provide a lot of these pixels, but 2 circular fisheyes taken with a 10mp camera just can't deliver the resolution (number of pixels) that 30 shots using a Multi-Row capture technique with the same camera.
 
There's lot of other factors, but more pixels from more source photos produces more resolution

Hi Harrison!
 
Harrison quote:
------------------
"Out of interest, is a shaved Nikon 10.5mm much better quality than my Sigma 8mm EX DG on the 5D with a lens adaptor"? I notice the Sigma 8mm is not as sharp as the Canon 15mm full frame fisheye so my full screen QTVR's lack resolution and detail - but are quicker to make.
 
I don't know... I've viewed some some excellent Sigma 8mm f/3.5 panos. The only credible comparisons I've seen are the 10.5 vs the Sigma 8mm f/4.0, where the 10.5 was clearly a superior lens... but the f/3.5 seems to be a far better lens than the f/4.0. The "shaved 10.5" simply allows the full expression of its native FOV, otherwise artifically truncated by the sunshade.
 
But shaved or unshaved the Nikkor 10.5mm has an inherent advantage over the Sigma 8mm, it has a bigger image circle (more pixels) to represent the equivalent FOV.
 
 
Which gets us to the second part of your question (Sigma 8mm vs Canon 15mm). I would expect the Canon to be sharper. Comparing apples to apples - what do the Canon 15mm's source images look like when they are reduced to show the same corresponding image area as the Sigma's? (i.e. the Canon image will be a subset of that shown by the Sigma.)
 
 
========
Some ideas for improving the Sigma 8mm output - if you haven't already tried them:
 
1) Shoot with f/8 - f/11 whenever possible; the sharpest settings for any lens.
 
2) Pay close attention to your focus; it is so easy to miss the setting given the huge depth of field of a fisheye (full circle especially). Shooting an indoor pano with objects close by (less than three feet) out to twenty feet will have a different optimum focus (depth of field ~ near/far focal distance; try googling hyperfocal distance) than that of panos where your subjects are six feet to infinity.
 
Experiment with these settings by having man-made angular objects (no fluttering leaves!) 1 foot; 3 feet; 8 feet; 20 feet; 40+ feet in your image and change the focus slightly in methodical increments. Then compare the images on your computer.
 
3) Use the DxO program ( http://www.dxo.com/intl/photo )
There is no better way to optimize your images than with this program - the deblurring feature has no equal.

Thanks Jim, appreciate your advice. Have tried all your suggestions apart from the DxO program - will look in to this.
In the meantime, I'm still playing around with the Canon 5D and my new Sigma 8mm EX DG - and having some teething problems.
I'm using the Nodal Ninja 3 (and their suggested settings) and Stitcher Unlimited 5.6.1 with three frames around (at 0 degrees) and one up for Zenith. Sometimes they all stitch ok (with minor errors) but more often only three out of the four shots get the "green" stitching ok symbol. However, if i shoot four around (not three) and one up, everything is hunky dory and they all stitch.
I've tried calibrating the fish eye lens etc, but seem to be going in circles and not getting anywhere.  
I want the images to be used for full screen QTVR - especially action shots - hence the preference for fewer frames.
Any suggestions how i can get all the images to stitch ok more of the time ?
Your valued advice would be very much appreciated and save me pulling any more of my hair out ... (
 

What would you suggest as the preferred fisheye for the full frame 5D's sensor (can you get away with 3 fisheye's)

Sigma 8mm f/3.5 or possibly the new Sigma 10mm f/2.8 I'll need to check the area of coverage with the 10mm on a full frame sensor.

trsdellis
According to "Canon's EF Lens Specifications" chart the Canon 15mm Fisheye will produce an image that will fill the whole Full Frame sensor on the 5D with a FOV of 180° diagonally.  
 
The traditional shoot technique for a lens like this is 6 shots around, plus a Zenith and Nadir the same as a Nikkor 10.5mm on an APS-C/DX sensor. With the 6 shots you're capturing more pixels so you can create a larger sphere/cube that can be displayed larger, even full screen.
 
You'll probably have more pixels than you'll need for full screen display with any of theses lenses
 
Here's a little spherical math that might explain; a good size for most spherical images (equirectangular projection) is 2048 x 1024 pixels (that would be a cube with 512x512 cube faces) . Its a size iPIX uses as one of its default sizes. The 2048 is what the circumference of the equator of the sphere will be. These display fine at 320x240 even 400x300 depending on the compression
 
Doubling the dimensions to 4096 x 2048 pixels puts the equator circumference at 4096 pixels (or the cube faces at 1024 x 1024). A typical Initial FOV setting for many viewers like QuickTime, PTViewer, even the iPIX Java Viewer is around 90° wide (one quarter of the equator). So on a 1024 x 768 monitor there's exactly enough (4096 divided by 4) pixels for the monitor width. I've alway found going a little larger on the image size and letting the video card do a little down-sampling produces a really nice image so I've been planning on using 5000 x 3750 for full screen display when I move up from my Coolpix setup.
 
The point I'm making is; because of the most common screen sizes (currently 1024x768 according to the statistics from my ad serving provider) creating anything a lot larger than 5000 x 3750 is just redundant prixels that won't be displayed unless the person viewing zooms really far in. Take note that the RealViz Forum is designed for a 1024 pixel wide monitor
 
So a 4 shots with a Sigma 8mm on a 10 megapixel camera in portrait postion will capture 10,368 pixels around the equator. With approximately 1/3 overlap that leaves about 6900 pixels around the equator. Not a lot more than the 5000 I suggested.
 
Taking 6 shots with a Nikkor 10.5mm on a 10 megapixel camera in portrait position will capture 15,552 pixels around the equator. With approximatly 1/3 overlap there will be about 10,357 pixels around the equator. A lot more pixels than needed for a "neutral zoom" but they would allow a lot of high resolution deep zooming.
 
Like I and others have said, if you want more resolution, you need more pixels captured. The way to do that is a lens with a narrower field of view so you have to take more photos around. A 55mm lens would need 12 to 16 shots around (probably 2 rows of them). You'd get 20,715 to 27,620 pixels after 1/3 overlap around the equator. That's over twice the resolution of a 10.5mm Fisheye on an APS-C 10 megapixel sensor.
 
The problem with "better resolution" is balancing the need for more pixels with the desire of  most virtual photographers to shoot fewer shots. And after some point, a really large sphere or cube file just has no benefit on a 1024 x 768 computer monitor
 
Hope this little math lesson is helpful

On Ken Turkowski's site there is a great set of conversion calculators for a variety of circumstances:
http://www.worldserver.com/turk/qu [...] nores.html
 
Ken was one of the original members of Apple's QTVR team.

Thanks for the math lesson -  
 
I understand the concept of more pixels = better resolution etc.  
 
I tend to work with the original 10+MP images in CS2, stitch, then choose a render size relating to what the end use of the pano is going to be (always web use, anyting from 400x300 up to the 1K + size depending on the published websites visitors screen resolution log's - ususally tends to be 1024 x 768 (60%, with only 10% + or - going higher than that, and 10% at the old 800x600 - using CRT's probably!)
 
You are exactly right with your 1024 x cube face calculation - I found via trial and error that at todays average resolutions that tends to be the best for full screen, but with real estate thats way too big as I find when questioning viewers of virtual tours, they dont often view in full screen (just what I've been told)
 
Thanks for the reply's and have a good week.

Price Update
 
Adorama.com already has the new Sigma 4.5mm Fisheye on their website for $899