天空вℓσи∂ @skyblond@qoto.org

Almost half of the year has passed, but I'm still not used to typing 2024...

Oh man, don't tempt me to order an illuminometer.

Start testing glow sticks. Today's two are identical in terms of light: 558nm ± 2nm. The 1 hour one is super bright but decayed super fast. It barely glows at 1 hour. But at the beginning, it measured 668 lx, reported by my phone, when placing the stick on the screen.

The 10 hours one is not that bright, but still good for reading. It feels much easier to see compared to the blue glow stick: the blue one is bright, but it makes my eyes exhausted. The yellow one is much better. Since human eyes are much more sensitive to green light, I would expect the green one to be the best. But I'm also curious about the white one.

And with the last two images, you can tell the lacking of blue makes the color very off. You can hardly see blue. I mean, that's expected, and I know this. But seeing with my own eyes is just amazing.

Bird, but furry

That's a weird looking bird, LOL

Last weekend, me: WTF why it stops working???

Also me, several hours later: I forget to uncomment the line where I commented out for debuging.

15 hours and still glowing. But at this time the light is super week, I don't think you can read with that. You can only tell something is glowing.

Praise the chemistry, again.

Guess why I need to force myself saving some money? Because I can't help buying random stuff.

This week is glow sticks from Japan, 2 USD per stick. It's pretty expensive in terms of being a toy (it's not a toy). But guess what, it's super cooool!

I paid 0.526USD/Hour on AWS and only get 2 FPS.

I rent a GPU on vast.ai for less than 0.5USD/Hour, and I got 12.5 FPS.

Today's lesson: Do not trust AWS.

I need an explanation...

"Boost"

I mean, the heavy lift is done by onnx and yolo, I just invoke them and copy files based on the result.

But still, it's exciting to see my code actually working. BTW I'm using kotlin/jvm. So no python nonsense.

YOLOv8 detect birds in video clips.

I think the main issue of this toot is I use the default 640x480 resolution, which makes the picture too small to distinguish the details. Running on 1080P natively does cost a lot of RAM and time, but it can tell it's a bird very quickly and confidently.

I need to go sleep...

OMG it's working!!!

ONNX runtime for java with kotlin.
Using FFmpeg to read videos.

OpenCV can give boxes when using CPU, but somehow the boxes are all zero if running on CUDA.

The ORT is much better than OpenCV. I love onnx :)

Don't know why but yolov8 gives me 0 bounding box with opencv dnn.

ok, it's really confident about that