Finally got most of my hardware here! So excited to begin work!
Thank you @sampoder for all your help with the process!
Wrote a really really long email on various project proposals 😭. Also investigating Flutter and building MacOS apps. 🤔
Not much project work today, I did homework instead so here's a pic of my tr's cat.
Even more architecture work today. Got dependency injection, configuration, and module hosting worked out. Now to figure out how to run them in sync with each other....
Scaffolding the code structure for my robot!
Got PyDNet exported to TFLite with the help of the creator! What a great guy! ❤️ 🎉
Soooo day 3 of holidays and I'm sick 😭. Don't worry, it ain't the c word.
Hardware party!!! 🎉 🎉
Learning about point set registration for fusing point cloud data over time.
Finally got point cloud generation working from depth maps, only took 132 attempts! 😭 😭 The final code isn't even that complex, the problem was that I had to keep checking to make sure my tools were working like they should be. The result is also a little skewed since I don't have the actual FOV camera measures for the source image but that will be available when I work with the hardware.
Now excuse me while I go rest...
Started working on depth map to point cloud projection using the normal PyDNet model while I wait on resolving issues with PyDNet TFLite.
Also wrote a proposal for grant funding to build an Arweave application that would help with problems I discovered while sourcing ML models for this project. Excited to see where that leads!
Schools finally out for the semester! 🎉 🎉
Messaged an author of PyDNet to seek their help in fixing the problems I am having with getting quality depth maps. Also studied how to do kinematic modelling for two-wheeled vehicles which I'll use to steer my robot entirely from its back wheels!
Doing some more study on autonomous driving. Looks like I'll be making occupancy grids from point cloud data!
Did some more research into post-processing the depth map from Pydnet. What a shame to see people not fully documenting their methods! 😭
On the bright side, it looks like I can run Pydnet without an ML accelerator albeit at a much lower FPS which should be fine for an MVP 🎉 .
Got Pydnet working for depth estimation on TFLite (not using DenseDepth anymore since the model was too large)! Need to now work on smoothing out the output like they have in their demos then moving onto generating point cloud data from these maps!
Just wrapped up test season, finally able to focus on SOM! 🎉 🎉
Starting work on converting DenseDepth to TensorFlow Lite for realtime, mobile ML inference!
Learning about working with point clouds and applying Kalman filters for autonomous vehicles!
