Sunrise Mountain detection review

Frames

9,371

3,576 dark, 5,795 usable · night/day split found at luma 83

Raw detections

3,319

32% of frames had one

Suppressed

3,306

3,306 static · 0 dark · 0 sky

Candidates

0.2% of usable frames

Read this first

3576 of 9371 frames were too dark to carry information — this camera has no infrared illumination — and were never sent to the model. MegaDetector reported 3319 detections across the 5795 that were. Scene filtering set aside 3306 of them: 3306 were boxes that recur at fixed pixel coordinates on a camera that never moves, 0 fell in frames too dark to carry information, and 0 sat above the treeline where nothing terrestrial appears. 13 survived, across 5795 usable frames.

Candidates · 13

Survived static-object, dark-frame and above-treeline filtering. These are the only crops worth a classifier's time.

Sat 22 Aug 20264 detectionspeak 0.51

detection crop — candidate **0.51** sunrise_mountain_20260823T033120Z.jpgSat 22 Aug 2026, 8:31:20 pm PDT2026-08-23 03:31:20 UTC 156×193px

Sun 23 Aug 20261 detectionpeak 0.45

Mon 24 Aug 20262 detectionspeak 0.82

Tue 25 Aug 20262 detectionspeak 0.39

Wed 26 Aug 20264 detectionspeak 0.60

Recurring locations

The camera never moves, so a box that keeps landing on the same pixels is scenery. Shaded rows were suppressed.

Box (x1,y1,x2,y2)	Frames	Detections	Max conf	Outcome
`[1750, 523, 1920, 1039]`	2755	2854	0.93	suppressed
`[1585, 521, 1918, 1043]`	100	100	0.71	suppressed
`[0, 2, 248, 1018]`	50	50	0.54	suppressed
`[0, 42, 122, 1019]`	37	37	0.48	suppressed
`[275, 556, 448, 695]`	34	34	0.69	suppressed
`[1751, 0, 1919, 1044]`	28	28	0.47	suppressed
`[1380, 267, 1628, 630]`	22	22	0.75	suppressed
`[374, 0, 697, 680]`	20	21	0.43	suppressed
`[1214, 734, 1303, 940]`	18	18	0.84	suppressed
`[913, 366, 1084, 683]`	12	12	0.48	suppressed
`[1611, 230, 1881, 605]`	10	12	0.47	suppressed
`[436, 342, 683, 697]`	10	10	0.69	suppressed
`[1858, 605, 1919, 1041]`	9	9	0.46	suppressed
`[1547, 0, 1918, 1040]`	7	7	0.70	suppressed
`[1183, 485, 1321, 660]`	7	7	0.61	suppressed
`[950, 535, 1064, 682]`	7	7	0.68	suppressed
`[1895, 533, 1920, 1039]`	6	6	0.37	suppressed
`[21, 12, 717, 688]`	6	6	0.38	suppressed
`[1110, 162, 1255, 311]`	6	6	0.63	suppressed
`[913, 366, 995, 684]`	6	6	0.61	suppressed

Suppressed · 3306

Kept and labelled rather than dropped, so the cost of the filtering stays auditable and a real animal caught in a suppressed cluster can still be found.

Show suppressed detections (first 120)

Fri 21 Aug 20265 detectionspeak 0.86

Sat 22 Aug 202661 detectionspeak 0.91

Sun 23 Aug 202610 detectionspeak 0.93

Mon 24 Aug 20262 detectionspeak 0.88

Tue 25 Aug 202610 detectionspeak 0.87

Wed 26 Aug 20268 detectionspeak 0.90

Thu 27 Aug 202617 detectionspeak 0.90

Fri 28 Aug 20267 detectionspeak 0.91

These pages, the code that produced them and this prose are written with AI coding agents under human direction, and reviewed before publication. How these reports are produced.