Yacht photo repository

[Nick Costen]

Hi Everyone,

I seem to remember that there used to be a website on which people posted pictures of boats they'd photographed, with the idea that the owner would then be able to download it. Does anyone know if it's still running, and if so, what's the URL? One of my students wants to do a research project on automatically reading the sail-numbers, so we need to find a good set of photos. A quick look at Google shows that almost all the time people frame the image to capture the hull, and crop the sail-numbers out.

Any ideas?

See you,

Nick

 

[matt1]

Did you mean this?

yourboatpix.co.uk is closing

 

[Nick Costen]

Thank you very much indeed - alas, yes I did. Oh well.

I'll see where else that sort of image might be available.

Nick

 

[anoccasionalyachtsman]

Hi Everyone,

I seem to remember that there used to be a website on which people posted pictures of boats they'd photographed, with the idea that the owner would then be able to download it. Does anyone know if it's still running, and if so, what's the URL? One of my students wants to do a research project on automatically reading the sail-numbers, so we need to find a good set of photos. A quick look at Google shows that almost all the time people frame the image to capture the hull, and crop the sail-numbers out.

Any ideas?

See you,

Nick

Why not get your student to approach the likes of Beken. They'd probably be very interested in a bit of software that could automate identification.

 

[Minerva]

How about trying some Facebook groups? A group for a racing dinghy class would likely have many photos with sail numbers prominently on display

 

[BabySharkDooDooDooDooDoo]

Sailing Scenes | Cruising Yachts – an A-Z

 

[Sandy]

There are a lot of sailing groups on Flickr that could be used.

 

[Rappey]

They'd probably be very interested in a bit of software that could automate identification.

Long ago Beken were often out in ribs photographing boats.. You could then go to their website and buy your pic..
Nowadays with digital and mobiles you just point your camera at a subject then can do a web search, point the camera at foreign script and it translates to whatever language you want live on your screen or translate words and numbers in a picture to text so im sure there is plenty out there like anpr that can automate identification..

 

[anoccasionalyachtsman]

Long ago Beken were often out in ribs photographing boats.. You could then go to their website and buy your pic..
Nowadays with digital and mobiles you just point your camera at a subject then can do a web search, point the camera at foreign script and it translates to whatever language you want live on your screen or translate words and numbers in a picture to text so im sure there is plenty out there like anpr that can automate identification..

I know all of that - I've bought Beken pictures, and a good friend was right in the forefront of recognising data in images, but this student might just help one of the small marine photography businesses to start using it - and get some really good experience of the real world.

 

[Nick Costen]

Hi Everyone,

Thank you very much indeed for the information and advice. That will be a great help.

I'm obviously aware of the general image analysis and classification systems; they generally work on Bag-of-Words principles. However, I'm not sure how well they specifically work on sail numbers and symbols, and its a more interesting problem than ANPR because of the distortions of the surface, wider range of fonts, locations and sequences, presence of reversed numbers on the other side of the sail (not something I thought of initially, as I sail a red-sailed gaffer) and finally use of non-alphanumeric symbols to identify brands of boat.

More significantly, I can't find any suitably labelled and balanced datasets of these images. This is the first requirement of research in any computer vision field; it's impossible to train and in particular test an algorithm without the basic resources. Such datasets really need to be public, as that allows comparison of algorithms on an agreed base-line rather than fitting individual algorithms to individual datasets. If a field takes off, you usually see a sort of ratchet-effect, as algorithms improve until they can completely solve datasets, leading to collection of new, more complex sets and development of new algorithms.

In the longer term, what I'd like to do is develop systems for tracking yachts, in particular to automate the race-officer tasks. Of course, I know there's little or no commercial value and that people do it for fun - I'm an academic looking for interesting tasks in a field I enjoy. Again, there is little or nothing in the academic literature on it, although I can find a few papers on recognizing fishing boats and tracking small commercial vessels in dock or inland waterway situations. This particular student is just the start; we want to develop a track record in it before we move on to more extensive problems.

See you,

Nick

 

[anoccasionalyachtsman]

Hi Everyone,

Thank you very much indeed for the information and advice. That will be a great help.

I'm obviously aware of the general image analysis and classification systems; they generally work on Bag-of-Words principles. However, I'm not sure how well they specifically work on sail numbers and symbols, and its a more interesting problem than ANPR because of the distortions of the surface, wider range of fonts, locations and sequences, presence of reversed numbers on the other side of the sail (not something I thought of initially, as I sail a red-sailed gaffer) and finally use of non-alphanumeric symbols to identify brands of boat.

More significantly, I can't find any suitably labelled and balanced datasets of these images. This is the first requirement of research in any computer vision field; it's impossible to train and in particular test an algorithm without the basic resources. Such datasets really need to be public, as that allows comparison of algorithms on an agreed base-line rather than fitting individual algorithms to individual datasets. If a field takes off, you usually see a sort of ratchet-effect, as algorithms improve until they can completely solve datasets, leading to collection of new, more complex sets and development of new algorithms.

In the longer term, what I'd like to do is develop systems for tracking yachts, in particular to automate the race-officer tasks. Of course, I know there's little or no commercial value and that people do it for fun - I'm an academic looking for interesting tasks in a field I enjoy. Again, there is little or nothing in the academic literature on it, although I can find a few papers on recognizing fishing boats and tracking small commercial vessels in dock or inland waterway situations. This particular student is just the start; we want to develop a track record in it before we move on to more extensive problems.

See you,

Nick

Automating the number spotting for RO's would be a blessing - but a heck of a challenge. It might even persuade us to improve the SI's a bit so that competitors present themselves in a more orderly fashion (and close enough that we can heckle the Solent-based IRC boat with blue numbers on his black sails - he knows who he is...).