5 ms·
How does the "measure with your iPhone" thing work? That seems to be the most error-prone part of the entire process, yet you claim very high accuracy? That wou
by CountHackulus 12y ago
How does the "measure with your iPhone" thing work? That seems to be the most error-prone part of the entire process, yet you claim very high accuracy? That would seem to imply that you can somehow get accurate measurements out of the camera?
- mpenn 12y agoMy cofounder and I spent over a year developing the measurement technology -- it's a ton of math / machine learning. The process is also easy to do (you don't need to stand as still as a statue or anything). We also really stand behind that 20% number. We brought 4 professional tailors in and 35 people, and we were 20% more accurate than the professional tailors. If anyone wants, I can go into more detail on the calculations of us vs. tailors.
- rhema 12y agoPlease, do tell. Is there a whitepaper or blog post somewhere about how this technology works? Also, do you suppose you could get better results from a consumer depth-camera, such as the Kinect?
- mpenn 12y agoNo white paper or blog post explaining the behind-the-scenes stuff (kept proprietary for now, but we may write something later). Yes, the Kinect would probably be more accurate (it was one of our original ideas), but it wouldn't have made a very good business. No one would have been able to easily use it in his home (not many people own a Kinect, 3rd party apps for xbox aren't big and when I have seen a Kinect, it doesn't capture your whole body). We found that our accuracy is more than sufficient for shirts (we've shipped a bunch already and gotten overwhelmingly positive feedback). Accuracy is just 1 aspect of a great looking shirt though; style is also really important, and we've spent a long time refining that as well.
- mpenn 12y agoRealized I didn't actually explain the measurement process. You put the phone down by leaning against a wall (or chair or something else), stand ~10 feet away (so your whole body is in frame) and do 1 spin. The software takes care of the rest. The app walks you through in much more detail how to do it. You also need to be in contour fitting clothing, so it's recommended to be at home.
- carbocation 12y agoAny reason to have it that way rather than shirtless? (My guess is that shirtless would be more accurate, but that you guys decided the privacy tradeoff was not worth it for the minimal accuracy gain.)
- trevyn 12y agoWhen you do it in the app, it does say that shirtless and briefs or boxer briefs are OK. I assume that fully nude would also be OK, but I guess they didn't want to go there. ;)
- mpenn 12y agoAs trevyn mentioned, we have both options -- shirtless or a contour fitting shirt. We've found there's no accuracy difference. You can't wear a normal, loose t-shirt or shirt; that unfortunately destroys our accuracy.
- rahimnathwani 12y agoI just tried it. The UX is truly amazing.
- mpenn 12y agoThanks! We have worked really hard to make the measurement process easy to do, testing with dozens of people for months.
- mpenn 12y agoOk, some more of the statistics on the 20% number. We had each tailor measure each person across 17 different measurements. Across those 4 tailors (on each person and each measurement), you could see a standard deviation. So then we also looked at the standard deviation of the average of those 4 tailors vs. the MTailor software's system. Our standard deviation was 20% lower. Happy to go into even more detail if people are interested!
- dloman 12y agoLooks like a really cool product. Are you thinking of moving to android? Don't mean to nit pick but you are mixing up accuracy and precision. If your standard of deviation is 20% lower, then you are more precise then professional tailors. Accuracy denotes some distance from some "true" value. Accuracy != Precision http://en.wikipedia.org/wiki/Accuracy_and_precision http://en.wikipedia.org/wiki/Accuracy_and_precision
- mpenn 12y agoDefinitely planning on Android (but not for a few months). You're definitely right about precision vs. accuracy. The phrase never polled as well. Maybe if we'd run our surveys in the statistics department it would have been different. :-)
- jrockway 12y agoRegardless of what polls tell you, calling precision accuracy is inaccurate.
- rahimnathwani 12y agoI'd love to know more detail, if you can share it without spilling the secret sauce.
- mpenn 12y agoFrom the video, we build a full 3D scan of you (lots of math / machine learning here). Then we essentially take a digital tape measure to your 3D scan.
- avtar 12y agoThis isn't iPhone related but perhaps others will find the project interesting: http://openfitlab.com http://openfitlab.com https://www.flickr.com/photos/kylemcdonald/9150428195/in/pool-2202673@N25 https://www.flickr.com/photos/kylemcdonald/9150428195/in/poo...