Note A4: synthetic occlusion scenes with exact ground truth counts, for scoring visibility corrected counts and calibrated interval coverage.

Note E1: synthetic multi camera transit scenarios for scoring cross camera identity linking precision and bounded memory footprint.

Note E4: synthetic kinematic trajectories for scoring how early an incident is predicted and whether each prediction was later fulfilled or falsified.

Note A3: synthetic scenes with known camera geometry for scoring self calibration accuracy from pedestrians walking through a static camera view.

Note A5: synthetic radiometric imagery for scoring the data engine and self supervised pretraining harness ahead of any GPU pretraining run.

Note B1: sample prompts and pipeline smoke tests for the language guided vision model factory, from a natural language task description to a calibrated exported model.

Note The Dhi Labs blog: versioned technical posts covering the honesty thesis and the product write ups, hosted as a dataset since Hugging Face has no public posts API.

Dhi-Technologies/prompt2model-reference-onnx

Image Classification • Updated 22 days ago

Note B1: the reference ONNX export the demo Space evaluates, paired with the example fixtures above.

Dhi-Technologies/thermal-mae-vits-pilot

Image Feature Extraction • Updated 22 days ago

Note A5: the LLVIP MAE pilot encoder checkpoint, paired with the thermal benchmark dataset above.