product-recognition-demo/recognization/data/external-eval/tier3-sources/eval_result.json
kiyo9w 962cdaab5f test(tier3): run the production pipeline over both reusable clips
The two clips were staged but never exercised. One frame per second through the
production detector and the shipped local lane:

- The 1080p supermarket pan: 227 crops, 0 accepted. Correct open-set behaviour -
  the frame is German pickle jars, not catalog products.
- The 360p mayonnaise bay: 1623 crops, 24 accepted (1.5%), across seven SKUs.
  The clip visibly contains Kewpie products, so those are not automatically
  false accepts, but at 640x360 the labels are too soft for per-SKU ground
  truth.

Recorded as an accept-rate probe with false_accepts and accuracy left null
rather than estimated. Tier 3 still needs an operator-recorded walk.
2026-09-29 03:39:33 +09:00

46 lines
1.5 KiB
JSON

{
"dataset": "tier3-sources",
"tier": "tier3_proxy_candidate",
"note": "Not Tier 3: one frame per second was extracted from two reusable public clips and run through the production detector plus LocalLaneAdapter. No per-facing SKU ground truth exists, so this is an open-world accept-rate probe, not an accuracy score.",
"method": {
"frame_sample_hz": 1,
"frame_long_side": 1280,
"detector": "production detector",
"identity": "stock-lopia-clean production local lane"
},
"clips": [
{
"file": "media/hanzomon-supermarket-pan-132-152.mp4",
"frames": 20,
"detected_crops": 227,
"accepted": 0,
"accept_rate": 0.0,
"interpretation": "Correct open-set behaviour. Pixel inspection of frame 10 shows German Kuhne pickle jars, not catalog products."
},
{
"file": "media/kat-in-japan-mayonnaise-aisle-pan-694-720.mp4",
"frames": 26,
"detected_crops": 1623,
"accepted": 24,
"accept_rate": 0.0148,
"accepted_by_sku": {
"sku_l05": 13,
"sku_l03": 4,
"sku_l04": 2,
"sku_l08": 2,
"sku_l01": 1,
"sku_l09": 1,
"sku_l10": 1
},
"interpretation": "The clip visibly contains Kewpie and adjacent mayonnaise products, so accepted crops are not automatically false accepts. At 640x360 the labels are too soft for exact per-SKU ground truth. The 1.5% accept rate is therefore a triage signal only; Tier 3 still needs an operator-recorded walk with labels."
}
],
"summary": {
"frames": 46,
"detected_crops": 1850,
"accepted": 24,
"accept_rate": 0.013,
"false_accepts": null,
"accuracy": null
}
}