Playing with MSR Cliplets

Cliplets is nifty, free tool from MSR that allows you to take a video and turn it into what appears to be a still photo. If you have the right scene, it results can be surprising and delightful.

Here’s my first go, say hi to Finney.

FinneyCliplet1

The picture above looks like a still photo, but then Finney pricks up his ear. It’s not a video, it’s an animated gif created with Cliplets.

About a year ago Jamie Beck and Kevin Burg made a splash by hand stitching images together to create moving still photos they called “cinemagraphs”. For an example of this technique done really well, check out a blog post from the Washington Post: Cinemagraphs: What it looks like when a photo moves.

Creating their cinemagraphs probably took Beck and Burg many hours of painstaking work in Photoshop. I don’t have that kind of time or patience. Fortunately, Microsoft Research Cliplets makes creating one of these still images quite simple. The UI is fairly intuitive, but there are a few things that need a little explaining. The best way to learn the features is to watch the short tutorials on the Cliplets page.

It’s worth noting, in order to produce a quality image you really do need good source material. The video I used for the picture of Finney was shot on my cellphone camera. The camera does shoot in 720p, but even so, the dark areas are grainy and parts of the video were blurry and couldn’t be used.

You know all those times you mistakenly had your camera set to video instead of picture? Time to go back and have some fun with those pictures.

Sometimes things just work out

My phone’s camera has a mode where it takes a 3×3 or 2×2 grid of images by snapping pictures in series. Press the shutter button and it starts going. There’s the typical camera phone lag to get started and then it takes pictures on its own schedule. I tried it on Finney and here’s what I got:

Autopilot for picture taking ain’t too bad. It just reinforces the same old “take lots of pictures, you’ll get something you like.”

Sneaking your dog into the office and other reasons to use speech recognition

Today is one of those rare days that I have no meetings.  And also happens to be a day that Paula needs to spend time getting ready for a house guest.  Finney is a good dog but he can be kind of needy.  To free up Paula I brought Finney into the office

For one complication was sneaking Finney into the office, however, is that when he is in new situations he gets nervous and whines a lot.  Now, sneaking a dog into the office doesn’t work very well if the dog is making a lot of noise.  The way to keep a nervous Finney quiet is to continuously pet him.  While petting him keeps him quiet, my hands are unable to do any typing.  Sounds like a good reason to try out speech recognition.

It may be my microphone, or it may just be that I need to do more training of the speech recognition, but my initial use has been slow going.  I’ve been using speech recognition to type this blog post, and I find that it is taking me three times as long as it would take if I were typing.  I can touch type and and fairly accustomed to putting my thoughts down directly from brain to fingers.  Part of the delay, I am finding, is expressing my thoughts in this new manner.  If I stare at the screen waiting for my words to appear, I’m brain just freezes.  There is a significant delay between voicing a word and having it appear on the screen.  For the touch typist who is used to seeing the output immediately on the screen it is distracting to have to pause ones stream of consciousness while waiting for the computer to interpret your words.

I also find if I speak too quickly the computer will run my words together to form similar sounding words.  This is usually frustrating as it requires frequent editing of the sentence is high and have just dictated.  Most commonly the errors are in the form of incorrect words but it gets even more frustrating when the dictation is incorrectly interpreted as commands to the program (it has tried several times to close this blog post prematurely).  It does, however, provide some amount of amusement when it makes errors like taking ” down directly” and turning it into ” downed rectally.”

As I struggle through, however, I find that through a combination of training the speech recognition engine and training Reeves, I am getting better at am using text to speech.  I wonder if I’ll ever get to the point where I can speak naturally to the computer and have it be acceptably accurate.