Dv04 commited on
Commit
3d5f565
·
1 Parent(s): 075475c

Experiment: remove index.html to test README.md markdown fallback for org card

Browse files
Files changed (1) hide show
  1. index.html +0 -170
index.html DELETED
@@ -1,170 +0,0 @@
1
- <!doctype html>
2
- <html>
3
- <head>
4
- <meta charset="utf-8" />
5
- <meta name="viewport" content="width=device-width" />
6
- <title>Dhi Labs: honest edge vision AI</title>
7
- <link rel="stylesheet" href="style.css" />
8
- </head>
9
- <body>
10
- <div align="center">
11
- <h1>Dhi Labs</h1>
12
- <p><strong>Edge native video analytics engineering, built to say "I don't know" instead of guessing.</strong></p>
13
- </div>
14
-
15
- <h2>Who we are</h2>
16
- <p>Dhi Technologies builds video analytics software that runs directly on Jetson class edge
17
- hardware, close to the camera, rather than shipping frames to a cloud model. The engineering
18
- identity is honesty as a feature: every component here quantifies its own uncertainty and is
19
- built to refuse an answer it cannot support instead of guessing.</p>
20
-
21
- <hr />
22
-
23
- <h2>Honesty as a feature</h2>
24
- <table>
25
- <tr>
26
- <td width="50%" valign="top">
27
- <h3>Refusal gates</h3>
28
- <p>Components decline to answer when the input falls outside their calibrated support,
29
- instead of emitting a confident guess.</p>
30
- </td>
31
- <td width="50%" valign="top">
32
- <h3>Calibrated confidence</h3>
33
- <p>Where a product reports an interval or a confidence score, the coverage of that
34
- interval against ground truth is itself measured and reported, including when it is
35
- imperfect.</p>
36
- </td>
37
- </tr>
38
- <tr>
39
- <td width="50%" valign="top">
40
- <h3>Falsification ledgers</h3>
41
- <p>Predictive components log what they predicted, what actually happened, and whether the
42
- prediction was fulfilled or falsified, rather than only surfacing hits.</p>
43
- </td>
44
- <td width="50%" valign="top">
45
- <h3>Stated limitations</h3>
46
- <p>Every dataset and demo states plainly whether its numbers are synthetic or real
47
- world, and what has not been checked yet. No claims of state of the art or foundational
48
- status, only what was actually measured.</p>
49
- </td>
50
- </tr>
51
- </table>
52
-
53
- <p>Two more things follow from the same policy: no SOTA claims, no customer names, and no
54
- deployment claims (this org is a public research and engineering surface, not a sales page),
55
- and every benchmark below is synthetic first and disclosed as such, so every claim can be
56
- checked exactly while real world validation is still in progress.</p>
57
-
58
- <hr />
59
-
60
- <h2>Products and evidence</h2>
61
- <p>Six products, each with a benchmark dataset built from procedurally generated ground truth
62
- and a live static demo Space you can exercise in the browser.</p>
63
-
64
- <table>
65
- <thead>
66
- <tr>
67
- <th align="left">Product</th>
68
- <th align="left">What it does</th>
69
- <th align="left">Demo</th>
70
- <th align="left">Benchmark dataset</th>
71
- </tr>
72
- </thead>
73
- <tbody>
74
- <tr>
75
- <td><strong>Amodal Counting</strong> (A4)</td>
76
- <td>Visibility corrected counting through crowds and occlusion: a calibrated interval
77
- instead of a bare point count.</td>
78
- <td><a href="https://huggingface.co/spaces/Dhi-Technologies/amodal-counting-demo">Space</a></td>
79
- <td><a href="https://huggingface.co/datasets/Dhi-Technologies/amodal-counting-benchmark">Dataset</a></td>
80
- </tr>
81
- <tr>
82
- <td><strong>Multicam Reasoning Memory</strong> (E1)</td>
83
- <td>Cross camera identity linking with transit time priors and weeks scale bounded
84
- memory; every answer carries provenance, and it refuses to link when unsure.</td>
85
- <td><a href="https://huggingface.co/spaces/Dhi-Technologies/multicam-reasoning-memory-demo">Space</a></td>
86
- <td><a href="https://huggingface.co/datasets/Dhi-Technologies/multicam-reasoning-memory-benchmark">Dataset</a></td>
87
- </tr>
88
- <tr>
89
- <td><strong>Causal Predictive Alerting</strong> (E4)</td>
90
- <td>Predicts an incident seconds before it happens from kinematic trajectories, and
91
- proves why via counterfactual replay and a fulfilled or falsified ledger.</td>
92
- <td><a href="https://huggingface.co/spaces/Dhi-Technologies/causal-predictive-alerting-demo">Space</a></td>
93
- <td><a href="https://huggingface.co/datasets/Dhi-Technologies/causal-predictive-alerting-benchmark">Dataset</a></td>
94
- </tr>
95
- <tr>
96
- <td><strong>Fixed Camera 3D</strong> (A3)</td>
97
- <td>Turns an ordinary fixed camera into a metric 3D sensor by self calibrating from
98
- people walking through the scene: no GPU, no model weights.</td>
99
- <td><a href="https://huggingface.co/spaces/Dhi-Technologies/fixed-camera-3d-demo">Space</a></td>
100
- <td><a href="https://huggingface.co/datasets/Dhi-Technologies/fixed-camera-3d-benchmark">Dataset</a></td>
101
- </tr>
102
- <tr>
103
- <td><strong>Thermal Perception</strong> (A5)</td>
104
- <td>A radiometric data engine and self supervised pretraining harness for thermal
105
- native perception, with CPU verifiable math ahead of any GPU pretraining run.</td>
106
- <td><a href="https://huggingface.co/spaces/Dhi-Technologies/thermal-perception-demo">Space</a></td>
107
- <td><a href="https://huggingface.co/datasets/Dhi-Technologies/thermal-perception-benchmark">Dataset</a></td>
108
- </tr>
109
- <tr>
110
- <td><strong>Prompt2Model</strong> (B1)</td>
111
- <td>A language guided vision model factory: prompt to dataset to trained model to a
112
- calibrated, quantized, ONNX exported artifact, with an accuracy floor refusal gate.</td>
113
- <td><a href="https://huggingface.co/spaces/Dhi-Technologies/prompt2model-demo">Space</a></td>
114
- <td><a href="https://huggingface.co/datasets/Dhi-Technologies/prompt2model-examples">Dataset</a></td>
115
- </tr>
116
- </tbody>
117
- </table>
118
-
119
- <p>Everything above, plus the blog dataset, is indexed in one place: the
120
- <a href="https://huggingface.co/collections/Dhi-Technologies/dhi-labs-honest-edge-vision-ai-6a4eb297cbd60f5f673cc2d7">Dhi Labs collection</a>.
121
- Browse <a href="https://huggingface.co/Dhi-Technologies?type=dataset">all datasets</a> or
122
- <a href="https://huggingface.co/Dhi-Technologies?type=space">all Spaces</a> directly.</p>
123
-
124
- <hr />
125
-
126
- <h2>Writing</h2>
127
- <p>Four technical posts, each grounded in the committed repo numbers. Hugging Face has no
128
- public Posts or Articles API, so these live as versioned markdown in the
129
- <a href="https://huggingface.co/datasets/Dhi-Technologies/blog">blog dataset</a>.</p>
130
- <ol>
131
- <li><a href="https://huggingface.co/datasets/Dhi-Technologies/blog/blob/main/01_ai_that_refuses_to_guess.md">AI that refuses to guess</a>: the thesis, calibration, refusal gates, provenance, and
132
- falsification ledgers as product features.</li>
133
- <li><a href="https://huggingface.co/datasets/Dhi-Technologies/blog/blob/main/02_e1_precision_first_linking.md">Precision first cross camera linking (E1)</a>: how a uniqueness guard lifted
134
- synthetic site precision from 0.918 to 1.0, and how weeks scale memory stays under 2 MB.</li>
135
- <li><a href="https://huggingface.co/datasets/Dhi-Technologies/blog/blob/main/03_a4_calibration_over_accuracy.md">When the error bar is the product (A4)</a>: calibration honesty, the analytic
136
- interval undercovered (0.50 to 0.63 versus a claimed 0.90); conformal calibration widened it
137
- 6.7x to 14.5x.</li>
138
- <li><a href="https://huggingface.co/datasets/Dhi-Technologies/blog/blob/main/04_portfolio_overview.md">Six products, one honesty thesis</a>: a portfolio overview tying all six products to the
139
- honest by construction thesis.</li>
140
- </ol>
141
-
142
- <hr />
143
-
144
- <h2>Research status</h2>
145
- <p>Research papers describing these methods are in preparation and have not yet been published
146
- on arXiv or any other venue. The two whitepapers on the labs page are internal architecture
147
- write ups, not peer reviewed papers, and are labeled as such.</p>
148
-
149
- <h2>Code and evidence</h2>
150
- <p>The code itself is proprietary and closed source permanently. These datasets, benchmarks,
151
- and demo Spaces are the published evidence; closed source code, open evidence. Prompt2Model is
152
- a deliberate exception with one public, MIT licensed release:
153
- <a href="https://github.com/DHI-Technologies-Inc/Prompt2Model-Language-Guided-Vision-Model-Factory">Prompt2Model, Language Guided Vision Model Factory</a>.
154
- The GitHub org below is an identity link, not a browsable destination for the other five
155
- products.</p>
156
-
157
- <hr />
158
-
159
- <h2>Links</h2>
160
- <table>
161
- <tr>
162
- <td align="center"><a href="https://dhi-tech.com">dhi-tech.com</a></td>
163
- <td align="center"><a href="https://dhi-tech.com/labs">dhi-tech.com/labs</a></td>
164
- <td align="center"><a href="https://github.com/DHI-Technologies-Inc">github.com/DHI-Technologies-Inc</a></td>
165
- <td align="center"><a href="https://github.com/DHI-Technologies-Inc/Prompt2Model-Language-Guided-Vision-Model-Factory">Prompt2Model repo</a></td>
166
- </tr>
167
- </table>
168
- <p align="center"><em>Partnership or access inquiries: <a href="https://dhi-tech.com">dhi-tech.com</a></em></p>
169
- </body>
170
- </html>