4 findings : silent NLF inference loop, per-frame allocations in
the Metal renderer, unguarded top-level torch import, hardcoded
DETRPose model size. Out of scope here, kept for future cleanup.
Banner on the Multi-HMR plan flipped from OBSOLETE back to
IMPLEMENTED after the NLF re-pivot (NLF TorchScript bundles
a CUDA-hardcoded YOLO detector incompatible with Mac CPU/MPS).
NLF plan stays in tree for the historical analysis.
- Implemented Hydra visualizations in `web_realart/public/hydra/app.js` and `index.html`.
- Created WebGL visualizer using Three.js and WebGPU in `web_realart/public/webgl/app.js` and `index.html`.
- Developed main landing page in `web_realart/public/index.html` to showcase different visualizations.
- Set up WebSocket server in `web_realart/server.js` to relay OSC messages to web clients.
- Added health check API endpoint for monitoring server status.
L'électron rare