Model results

COCO

Choose a shared evaluation protocol to compare model scores. Versions, sources and configurations stay separate.

Atlas snapshot captured 10/19/2026, 10:20:28 AM UTC. This is an import timestamp, not the evaluation date. Coverage may be incomplete.

Reviewed recommendation and limits

COCO defines task-specific evaluation for boxes, instance masks and human keypoints. Read each task separately; AP is an aggregate precision metric, not classification accuracy.

Box and mask AP use intersection-over-union thresholds. Keypoint AP uses object keypoint similarity. Keep task, dataset edition, split, training regime and evaluation settings separate.. Imported groups below may use different protocols; this review does not certify each run. A shared dataset name does not establish comparable runs. The imported EdgeCrafter claims lack verified evaluator revisions, executed configurations, evaluated counts and measurement dates. They remain attributed publisher results without positions.

Publisher results COCO detection evaluation COCO keypoint evaluation

← All evaluation groups

COCO2017 box AP50:95

Source: EdgeCrafter (reviewed publisher results)
Version: 2017-validation
Measurement date: Not reported
Metric: AP · Higher is better
Evaluation scope: Publisher-reported object detection AP50:95 on COCO2017 validation, input size 640. Training regime Objects365 pretraining followed by COCO downstream training. Exact linked checkpoint https://github.com/capsule2077/edgecrafter/releases/download/edgecrafterv1_o365/ecdet_s_o3652coco.pth. The README attributes this value to this checkpoint; evaluated binary digest, exact evaluator revision, executed configuration, evaluated image or instance count, repeated runs and measurement date are not supplied. AP uses its task-specific aggregate, not an accuracy numerator divided by an example count. Latency configuration does not establish quality evaluation precision. This is an attributed publisher claim, not independently reproduced or a cross-publisher comparable run.

Positions are withheld because at least one result lacks explicit independent provenance or a documented evaluation scope. Missing metadata is shown as unreported; matching imported metadata is not independent verification of the protocol.

ECDet-S O365-to-COCO
intellindust/ecdet-s-o3652coco
53.6 AP
View modelEvidence model-card-reported