Rien has been building Google's AIY Vision kit. It's a Raspberry Pi Zero with camera, light, buzzer, and button that uses Tensorflow to do machine learning. After some issues getting the little beast to connect to the camera, everything now works and she's starting to experiment with some of the built-in machine learning models.
We first tried the object-detection model, which attempts to identify "things" in an incredibly busy scene. We had some successes (including correctly identifying an orange), although it kept wanting to identify Paul as "dough". No kidding. Is that a win or a fail?
You can't tell what features the Tensorflow model is using to make decisions about the world, but we were able to make some guesses. The curtains, pulled back and tied with an elastic, kept getting identified as a balloon, and yes, we could see the similarities. The orange, when squashed, got identified as a squash or a lemon or a ping pong ball. The lighting made a big difference and the background made a big difference. Of course, it is a stunningly hard problem to identify random objects in a crowded scene, so the fact that we could tag anything was a win. Next up is to build a little stage so that we can control the background and lighting a lot better.
The eventual goal is to use machine learning to make decisions about what is and isn't recycling, so next up is to build a little stage where we can control the background and lighting a lot better. This will simulate a recycling sorting environment.
We also used the AIY to detect faces. This turned out to be a lot easier than arbitrary model detection. The model worked pretty well, including successfully detecting The Joker, side profiles, faces in pictures, etc but NOT emojiis, dogs, etc.
Oh, and we printed a nice little box in PET-G to keep all the electronics safe!
We have the Raspberry Pi connected to its own monitor.
The camera is looking at a computer mouse.
This image gets correctly identified as being about a computer mouse.
Our makeshift orange stage.
Correctly guessed orange - only with 0.35 probability though, so fairly unsure.
The guess is "lemon". Makes some sense - the orange looks pretty squashed in this view.
Got all the faces in this photo.
Our nice PET-G box.
Tricky to get the parts nicely tucked away.
Rebuilding the setup inside of the box.
- LGMS 20 Time
- MMM The Event
We found out that CCHS Principal Chris Rogers was a big fan of the Dune series of books. Our favorite 3D modeller has a nice model of the sandworms of Arrakis, so we 3D printed it and gave it to him as a scary desk ornament. Look at those grim teeth, ready to chew up any wrongdoers sent to the principals office, or at least give them a nasty hangnail.
Look at those nasty teeth.
Sandworm of Arrakis bursting out of a desk.
- CHAOSMakers CCHS
We recently downloaded a free copy of Agisoft's photogrammetry software and bashed through the included examples. After a trying a few other small test projects where everything worked well, we were flush with success and clearly overconfident. We decided to try something a lot bigger: a complete photogrammetric reconstruction of the CHAOS Lab.
We had vague plans of reconstructing the place in 3D and then importing it into Unity so we could put it in VR. "And then dinosaurs would jump out!". Of course we had no idea...
We started by taking 1300 pictures of the CHAOS Lab. Initially, we methodically moved the tripod by a few inches, took a picture, and repeated. But after a couple complete cycles of the room (at different heights) our energy started to flag. To make things more interesting, we switched to free-form shooting of all the dark corners we thought we'd missed.
We then started grinding away on the photogrammetric process: aligning the cameras, generating the tie points, generating a dense point cloud, generating a mesh, and then mapping textures. Photogrammetry is a GPU-intensive process, and our first attempt bricked the lab computer so hard and for so long that we eventually had to move the process to an alternate CHAOS secret lair so we could let things run undisturbed.
Of course, we didn't quite understand all the tweakable parameters, so there were several iterations. But after a good week or so, we had something. It was a little ugly and not useable for our purposes, but we learned a LOT about the process. We are keen to try again....and in fact we had much better results here: Photogrammetric Reconstruction of a Car Bumper
The CHAOS Lab 3D model from outside. The big hole in the roof is where we were unable to match.
This is what "1300 pictures later" looks like.
This is a reconstructed 3D surface. Note big holes in feature-less areas, like the blue divider.
The matching surface reconstruction.
Another reconstructed view. The popsicle stick structures on the bookshelf didn't make it.
Equivalent 3D model reconstruction.
A confusing view of the estimated locations of the camera during picture taking. The nice rows were taken on tripod, and then we took pictures everywhere.
Sparse point cloud built by matching features in adjacent pictures.
Dense point cloud. Look how noisy it is!
Reconstructed 3d surfaces. We were surprised how the floors recovered better than the tables. Maybe because dirty floors have more features to match?
Another view of the room. The floor came in pretty well but the stuff on top of the bookshelf was confused.
The matching 3d model.
- CHAOSMakers LGMS
A few weeks ago we saw an inquiry on a local community Facebook group about the availability of 3D printing services in Canmore. We're pretty much fanatics about pushing plastic in CHAOSMakers, so we got in touch to see how we could help. One interesting discussion led to another and before you know it we were outside at night, in the snow, taking hundreds of pictures of a car bumper.
The general idea was to reconstruct, using photogrammetry, the 3D shape of a custom bumper to see if it would be feasible to 3D print a matching duplicate bumper using an industrial 3D printer. None of us really knew what we were doing, but we all worked together and later on in the CHAOSMaker secret lair we were able to come up with a fairly nice 3D reconstruction of the front bumper face.
We realized we would be unable to accurately reconstruct the bumper backside model without an original scan of the unmodified car. However, we were pretty pleased with the accuracy of the frontside model, especially given the dark & snowy conditions of the original photoshoot.
The final result. Not bad for shooting the original pictures at night.
A little closer view of the bumper. The green is physical tape, and the faint sphere is our 3D navigation icon.
- CHAOSMakers LGMS
Okay, okay, so it's an overly dramatic showy title! Our plan is to use a silicone mould-making kit to build a mould to churn out sugar-glass beer bottles - the kind that actors smash on tables and heads during rowdy fight scenes. We plan to make a couple of these bottles, test them out, and then stage a dramatic "bar fight" (with the full knowledge and consent of everyone of course). A little editing of our fake beer bottle smash and we're well on our way to become famous YouTube stars...
But really, the point of this silly exercise is to learn about silicone mouldmaking. Here's a few pictures of our first attempt...
2019-01-18 - Making Sugar Glass Bottles
We're making sugar glass bottles.
Prepping the beer bottle and greasing up the mold tube.
Dumping one part of the silicone out. We need to mix 2 parts to get it to work.
The sealing stuff is kind of goopy.
Using a headlamp to make sure the bottle is properly aligned.
Mixing both parts of the silicone together.
Carefully pouring it into the mold.
The dramatic long pour helps remove bubbles.
Covering up the top of the bottle.
Experimenting with some leftover silicone.
We had to leave the orange blob to set over the weekend. More later.
- CHAOSMakers LGMS
Welcome to Mountain Maker Madness. This is a prototype, experimental website meant to tell the stories of the some of the builders, creators, makers, teachers and innovators in Canmore and Banff - starting with students in the CRPS school system. We're not sure where this is going, and it certainly isn't ready for public viewing yet. But we're launching anyway, in the time-honored approach of "winging it".
To get us going, here's a 3D print of the Gates of Moria, from the "Lord of the Rings". If you're familiar with the series, the Gates only opened after a password was spoken - the Elvish word for "friend". Let's start this weird experiment by inviting you - our friends - into the gates of this website.
The print was printed at LGMS on a PRUSA mk3 and was painted by Colin F in LGMS Grade 8. It is from our favourite 3D modeller.