Datasets:

Dhi-Technologies
/

fixed-camera-3d-benchmark

Size:

Tags:

License:

Dataset card Files Files and versions

xet

Community

Acknowledge the terms to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Access is granted for non-commercial research and evaluation use only. By requesting access you agree not to redistribute this dataset or any derivative data, to cite Dhi Technologies in any publication or output that uses it, and to obtain a separate commercial license via dhi-tech.com before any commercial use. Access requests are reviewed manually by Dhi Technologies.

Fixed-Camera 3D Benchmark

Product: fixed-camera-3d ("fixedcam3d"): self-calibrates a fixed camera from ordinary walking people and then reports metric position. Pure Python (numpy/scipy), no GPU, no model weights.

Paper: Dhi Labs paper 08, Pedestrian-Only Self-Calibration with an Ill-Conditioning Certificate (manuscript in preparation).

Results status. The headline results below are taken from the paper 08 abstract. The full set of results, tables and per-condition breakdowns will be posted here when the manuscript is final.

What this dataset is

Procedurally generated scenes from known pinhole cameras, with walkers of known height and position, and detector-like box noise added. The self-calibration and localization code is the product code run on this synthetic input. The files in this dataset are synthetic.

What's in this dataset

scenes.jsonl (288 rows, about 6.9 MB): one row per camera configuration of an earlier grid over mount height, tilt, observation noise and generation seed. Each row has camera_truth (height_m, tilt_deg, focal_px, image_width, image_height), noise_px, seed, calibration_observations (80 noisy person-detection boxes with true_xy_m and true_height_m), and eval_observations (60 held-out boxes, empty where calibration did not converge).
bench_results.json (about 263 KB): grid_axes, the per-configuration grid table, and an aggregate block from the same earlier grid.

Scope of these files. The files come from an earlier grid release. The paper 08 results come from its own runs (1,020 installations over five seeds, with the sweeps described below). They are not recomputed from these files.

Results (paper 08 abstract)

Frozen validation gate. Combining head residuals, a tilt perturbation probe and fitting-half disagreement, the gate reaches ROC AUC 0.851 on 126 untouched synthetic mounts.
Position radii. Independently calibrated position radii cover at 0.901 against a 0.90 nominal level.
Acceptance. The gate accepts 72 percent of installations, with position error 0.35 m among accepted installations. At acceptance 0.26, for a cross-fitted target of 0.10, the error is 0.09 m.
Sweeps. Sweeps over focal length, distortion, roll, ground slope, crowding, truncation and person height (1,020 installations, five seeds) map where error and refusal change.
Public multi-camera check. On seven cameras of a public pedestrian dataset, a learned focal-length estimate reduces median error from 1.35 to 0.27 m. Keypoint witnesses from three pose models give a finite vanishing point.
Gauges and planning. The paper derives exact scale and yaw gauges and the minimum survey anchors for a planar similarity gauge. Exact planning values are 29 and 59 accepted installations for 90 and 95 percent accuracy at 95 percent confidence.
Drift. An offline moving-mount exercise retains drift refusal after signed-artifact rollback.

How to load it

import json
from huggingface_hub import hf_hub_download

repo_id = "Dhi-Technologies/fixed-camera-3d-benchmark"
scenes_path = hf_hub_download(repo_id, "scenes.jsonl", repo_type="dataset")
bench_path = hf_hub_download(repo_id, "bench_results.json", repo_type="dataset")

scenes = [json.loads(line) for line in open(scenes_path)]
bench = json.load(open(bench_path))

print(len(scenes), "scenes")
print(list(scenes[0].keys()))

Limitations

The files in this dataset are synthetic. The public multi-camera check is described in paper 08 and is not part of these files.
The paper 08 results depend on the stated sweeps and the gate settings in the manuscript.

License

This dataset is released under CC BY-NC 4.0 (non-commercial). Access is gated and requires manual approval. It is provided for non-commercial research and evaluation only. Redistribution is not permitted, and any publication or output using it should cite Dhi Technologies. Commercial use requires a separate agreement; contact dhi-tech.com.