these pretty traces!

thing is, golden handcuffs cost about $40k and are incredibly malleable, you can just bend them right off

I don't really want to pay full price for this government come April 15.

$15 wifi

sexy airline pneumatic tube audio

just living the dream around here

The Kansas City hackerspace used to be in caves under the city. The entrance was a just a lone elevator shaft in a parking lot.

If It Makes You Happy

Came for the name, stayed for the feels.

open.spotify.com/track/6k7H7h...

I set off the fire alarm so many times today.

it worked in real life!

- Frontier scan to detected a feed tape
- YOLO bounding boxes for all parts (model tuned on LumenPnP cams)
- Ran a capture at each detected part
- Re-ran YOLO to get the nearest bounding box
- Calculated the offset, reran capture

cc @thea.codes

This is so exciting.

My focus has been tray feeders for very low volume, and was wondering about rethinking feeders. Rather than a mechanical, dead reckoning approach (feeder start + n * 0.2mm), more like "I know where every part is on the build plate absolutely and will give you coords"

Woot! It mathed!

Yeah I used a picture with a few 0805 caps.

deduped, with highest confidence score. I'm going to guess that there's an offset, but probably a constant one

A map of all the components on the board, by mm. The overlap is a little disappointing, guessing an affine transform thing.

Yellower color is a confidence score weighted by distance to the center of the image.

Direct YOLO -> Pixel Calibration

scale_x = 0.01219138 mm/px
scale_y = 0.01153833 mm/px

idk, maybe it's right?... ish?

@thea.codes @opulo.io

It's taken a while to realize what's frustrating to me about 'vacations'.

Why are vacations like... food?

My ideal vacation is a huge crafting table, workshop, and library. It can still be with people! but like working on independent or joint projects, learning new things, etc.

Food? Cooking!

Ohh sooo pretty! 🥹

I haven't looked at part orientation, more focused on feeder setup.

Yeah! I'll get it cleaned up and shared in a bit. Right now working on bounding boxes with YOLO, but frustratingly separated from my equipment for a few days :(

"Write down 16 unusual facts about yourself" feels like a new low for ice breakers.

"Two Truths and a Lie" is the laziest icebreaker.

Pretty good sim-to-real! A few misses, but mostly identified regions that contain parts with the @opulo.io LumenPnP.

Spent most of the day researching moving to Estonia?... That can't be right...

Emergency Stop? We don't need that kind of negativity.

And combined with the image splits, did a frontier scan to figure out regions of the build plate that look promising.

It's only step 1, but it significantly cuts down the search space for automatically configuring feeders.

Did a simulated grid scan over the entire build plate to see regions that might be worth exploring more.

From here, built a simulator in a colab to stitch together images from any position. Messy, weird projection, stuff, but okay for this purpose. I split images into 9x9 grids and ran embeddings on these splits. This tells us "looks like there might be more parts top-left from here"

Been having a blast with the @opulo.io LumenPnp. But my overall goal is to throw a bunch of junk a build plate and assemble a circuit.

First step, I wrote a plugin to do a 1cm x 1cm scan of the entire build plate. Collected images, got embeddings, manually classified them.

cooking up some raspberry pi