Local, Memory, and World
Three public routes show three different parts of Praxis without a drone or an account. This page records what each route does and what it does not prove.
Praxis Local#
Local opens a selected video as a browser Blob URL. Frames are read into a canvas and passed to an ONNX session in the tab. The route attempts WebGPU first and falls back to WASM. It prints the backend that was actually created.
The browser path scales the frame to a maximum working edge of 960 pixels, reads overlapping 384 pixel tiles with 96 pixels of overlap, and merges overlaps per class. The opening highway clip is a prepared derivative of Ram Sherwani's Pexels video. Its saved result index is labeled as prepared and uses red. A current browser pass uses green.
Local boundary
The upstream LibreYOLO VisDrone preview card describes five training epochs and publishes no validation metrics. Praxis does not turn the sample into an accuracy claim. The no-upload statement applies to the selected media. The page still fetches normal application files, including the model and browser runtime.
Praxis Memory#
Memory uses Scenario 2 from the public NANCY multi-perspective traffic recording. The three web derivatives begin at raw source positions 42.25, 39.90, and 36.45 seconds. Those starts were chosen by hand around the staged fall. After trimming, the derivatives use zero playback offsets on one 11.5 second clock.
One participant was seeded manually in each view and followed within that view with OpenCV CSRT. OSNet x0.25 compares three prepared crops at the 2.0 second case moment. The displayed scores stay fixed so they cannot be mistaken for current-time inference.
Memory boundary
This case does not estimate source offsets, associate people across cameras, or identify a person automatically. The OSNet score is a case-relative appearance comparison, not biometric identification. The vehicle candidate ends when the participant leaves that view.
Praxis World#
World uses Open Forest Observatory mission 000173. The published source mesh contains 4,217,056 vertices and 8,430,086 faces. Praxis ships a uniformly sampled and transformed browser derivative with 263,566 colored points, plus a path through 800 recorded camera positions.
The four inset images are published source examples. They cycle independently and are not presented as matches to the selected position. The route, source page, license, and derivative restriction remain visible together.
World boundary
Praxis did not create the published reconstruction. The route displays positions and their connecting path, not camera orientation, view direction, or frusta. Mission 000173 is used for evaluation and visualization only, not model training.
Source records#
- Local sample: Ram Sherwani, Drone shot of highway with traffic, Pexels License.
- Memory case: NANCY multi-perspective traffic video recording, Scenario 2, CC BY 4.0.
- World source: Open Forest Observatory mission 000173, CC BY 4.0.