Content pfp
Content
@
https://warpcast.com/~/channel/faust
0 reply
0 recast
0 reaction
July pfp
July
@july
1/n: A Deep Dive on The software side of how we've tackled Camera Pipeline with Roc Camera: - We've got down a lot of paths exploring how to camera in Roc Camera from the software side - First, from an architectural decision for v1, we decided to go with Embedded Linux environment instead of FPGAs or microcontrollers (because we didn't have the RAM to do ZKPs, or didn't want to support multiple MCUs that had to be tied via Ethernet, which would require packet switching or I2C or SPI or something else) this also meant that we could pipe the camera feed and its data directly to the embedded compute itself - We did also think about the fact that we'd have to prove the camera feed - and how we might need a signer on the CMOS - gets complicated - With Embedded Linux, we ended up selecting CSI Camera + RPi4 - Embedded linux offers us the flexibility of using existing infrastructure so we don't have to build sw from scratch, which is (hypothetically nice) The SW arch of what we're doing on the camera is next:
1 reply
0 recast
10 reactions
July pfp
July
@july
2/n: Software arch / setup - There are 3 main parts to the camera software infrastructure we're building: the GUI framework, the backend infrastructure, and the camera pipeline - I'm going to focus mostly on these parts here: - Our proof of concept / prototypes used PyQt as both the GUI framework and hardcoded most of the logic into it as well, and we used the OpenCV out of the box as the camera pipeline - OpenCV though is bloated, and we didn't really use computer vision infrastructure from the get go, so it was overkill even though they do have good infrastructure to get the video directly from the camera at `/dev/device/video0` for example - It was quick to build, written in python and easy to do - but also had a lot of bugs, hard to extend etc so we started looking around for other options - We eventually settled on Flutter as the GUI, our own Rust coordination backend that we wrote (that we call Geist), and a camera pipeline that we started experimenting Next is more about experimentation, and paths:
1 reply
0 recast
4 reactions