
Thanks to Tara Hunt I was able to get into the Alpha II of Riya, a photo searching/recognition service.
The alpha registration is simple. Just follow the link, and get the Riya uploader tool. As mentioned that the uploading is a slow time consuming process, it certainly is true. So I just started off with a small set of photos (< 100, to just see what I get). It looked like that the 'Start upload' actually launches a javaw.exe process and connects to the photo uploading site (where your photos get pushed). This is certainly a memory hogging process (where I saw around 40 threads increased everytime I started upload). Not sure what else the uploader does (apart from sending a acknowledge to a mailer when upload is finished, so you can get a mail back), but since my set was small, it was done in under 1 hour.
Once the photos were halfway uploading, I logged onto the Riya.com site and saw an album come up with my photos coming in. One of the best first user experiences in a long time, I saw small rectangles around the faces in my photos. What do most of the photos have? People. Faces. Riya does a good job in finding faces (and recognizing them to some level). Though sometimes I also gave me a small rectangle (like the ones for a face) on some parts of the photo, where there was no face. These areas instead had artifacts which the software might confuse with eyes (necklaces kinda things).
Once you have started afresh, Riya will allow you to train it by using the drop down around the faces and adding names to the faces. The more you do this, more trained the service will be to recognize this person in future. I did train for a couple of people in couple of photos, and immediately started to see Riya recognizing to some extent. This was low in my case, as I skipped on training to the limit. I was skimming on the training part...to see more action.
Once you go back to the home page under your login, you can see the thumbnail representation of the images in your album(s). I must say that the thumbnails were not of a good quality, this might be due to the scaling on the browser side. Below the thumbnails you can see the name of people (in your photos) and a number (denoting the # of occurences a person was recognized). Clicking on a number you can see all the photos of that person, which could be neat once you have achieved maximum recognition of your photo library.
One pain point in training was as you start adding people, the drop down above the faces becomes long, and you have to scroll a lot to tag an existing name. Also once the layer of names shows up, there is no way to toggle this layer off, which becomes irritating after some time. The only way to close this layer is by clicking the dropdown itself again, which is normal windows behaviour, but if you have moved your mouse over to some other portion of the image (with the contacts layer showing), you have lost the drop down itself. Hope the team understands what I am saying here.
You can manually tag the photos from a bottom control bar, which appears when you mouse-over an photo. I noticed 2 small issues here.
1.Once you type in a tag, and click add, the textbox doesnt clear itself. You need to manually delete the text to add new tag.
2.I could not get a way to delete a tag, that was added from the bottom control bar.
You can manually tag objects / text in an photograph. This can be done by dragging a selection rectangle around the object/text to tag, and then name that selection. This could be used for frequently appearing text/objects in your photo library.
Given the alpha state, I was impressed by the recognition capability. At the same time given the apha state, I hope the user experience would go up in coming releases. The thumbnail quality is low to poor.
Technorati tags: riya,soalphaithurts,riyarocks
Sphere: Related Content