WitneyWW's picture
add code (configs, package, eval, scripts, README)
b1b6f13 verified
Raw History Blame Contribute Delete
1.26 kB
{
"task": "validation",
"calibration_id": "sept-09",
"created": "2026-09-09",
"method": "PNP (0.70 px middle)",
"applies_to_dates": [
"2026-09-09"
],
"rmse_unit": "px",
"cameras": {
"left": {
"camera_serial": "104122062574",
"rmse": 0.5820335045117107,
"created_at": "2026-09-09T07:41:39"
},
"middle": {
"camera_serial": "217222066989",
"rmse": 0.6957969799496961,
"created_at": "2026-09-09T07:41:39"
},
"right": {
"camera_serial": "143322063538",
"rmse": 0.6641288760028174,
"created_at": "2026-09-09T07:41:39"
}
},
"projection_chain": "sensor_pose (mocap mm) -> T_mocap_to_cam -> camera intrinsics -> pixel; gel center via T_gel_to_rigid",
"note": "Solved on 2026-09-09 at 07:41, a few hours after the recordings it is applied to. The recorder's live overlay ran on the June-26 epoch during collection, and projecting the GelSight centres into a mid-episode frame puts June-26 on the sensors while this epoch sits 20-40 px away; both solves fit their own points to 0.5-0.7 px and share an identical gel-to-rigid transform, so they disagree only about where the cameras were. This epoch is the operator's choice. The June-26 files are in the repository as calibration/epoch_2026-06-26 if you want to compare."
}