Dataset card: regenerated reports; the Nemotron alternative
Browse files
README.md
CHANGED
|
@@ -29,3 +29,38 @@ The rest is supporting evidence: `verify/` (fall-alarm verdicts and clips),
|
|
| 29 |
input). Face-recognition data is not published. Each room-1 report compares
|
| 30 |
its routine with the earlier dates (day 1, the first, has none); day 19
|
| 31 |
(room 2) has no baseline.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 29 |
input). Face-recognition data is not published. Each room-1 report compares
|
| 30 |
its routine with the earlier dates (day 1, the first, has none); day 19
|
| 31 |
(room 2) has no baseline.
|
| 32 |
+
|
| 33 |
+
The reports were regenerated on 2026-10-01 with the current report code
|
| 34 |
+
(GitHub `feature/identity-linker`, commit `d4d52a1`: fall times read on the
|
| 35 |
+
right clock, visit edges fitted to the detector, the get-up time after a
|
| 36 |
+
fall, company the detector does not show removed, and other fixes). Only the
|
| 37 |
+
language-model summary and report were re-run: the detector output, the
|
| 38 |
+
fall-alarm verdicts (re-run: the same answers), the clips and the VLM segment
|
| 39 |
+
summaries are the earlier run's, except that `verify/verdicts.json` now gives each confirmed fall's
|
| 40 |
+
time on the video's own clock (`fall_time_absolute`). The report-generation
|
| 41 |
+
times above are the earlier full runs'.
|
| 42 |
+
|
| 43 |
+
## Alternative segment-summary VLM: Nemotron
|
| 44 |
+
|
| 45 |
+
`results_nemotron/<video name>/` holds the same six days with Nemotron 3 Nano
|
| 46 |
+
Omni 30B-A3B Reasoning (NVFP4) instead of Qwen as the segment-summary VLM
|
| 47 |
+
and as the fall verifier; the language-model summary and report are the
|
| 48 |
+
same Qwen3-8B. Made with GitHub branch `feature/vlm-variants` (commit
|
| 49 |
+
`6bb2a45`), the stages `VERIFIER=nemotron MAP_VLM=nemotron bash
|
| 50 |
+
bash/run_video_system.sh` runs, on the same detector output, alarm events and
|
| 51 |
+
face-recognition data as `results/`. Each
|
| 52 |
+
folder has `json_summary/` (the report and the VLM's per-segment answers),
|
| 53 |
+
`verify/` (the alarm events and Nemotron's verdicts) and
|
| 54 |
+
`action_classes.json`; detector files and clips are those of `results/`.
|
| 55 |
+
|
| 56 |
+
| Six days | Qwen (`results/`) | Nemotron (`results_nemotron/`) |
|
| 57 |
+
|---|---|---|
|
| 58 |
+
| Real falls confirmed (of 2) / false falls | 2 / 0 | 2 / 0 |
|
| 59 |
+
| Real visitors found (of 14) / extra people listed | 14 / 0 | 14 / 4 |
|
| 60 |
+
| Resident's time in the room, mean error per day | 3.1 % | 5.8 % |
|
| 61 |
+
| Resident's glasses described right (days) | 6 / 6 | 4 / 6 |
|
| 62 |
+
|
| 63 |
+
Nemotron describes one person's clothes, glasses or hair differently from
|
| 64 |
+
one 10-minute segment to the next, so a visitor can be listed twice
|
| 65 |
+
(days 1, 6, 19) and part of the resident's day can appear as a visitor
|
| 66 |
+
(day 8). Qwen remains the default.
|