Skip to content

Latest commit

 

History

History
34 lines (25 loc) · 4.81 KB

SCENE_GRAPH_MODEL_ZOO.md

File metadata and controls

34 lines (25 loc) · 4.81 KB

Model Zoo and Baselines

Openimages V5 Visual Relation Detection

TODO add yaml file link

All the following models are inferenced using unconstraint method, the detection part use X152FPN pretrained on OpenImages, COCO, Visual Genome and Object365 dataset.

model recall@50 wmAP(Triplet) mAP(Triplet) wmAP(Phrase) mAP(Phrase) Triplet proposal recall Phrase proposal recall model config
IMP, no bias 71.64 30.56 36.47 32.90 40.61 72.57 75.87 link link
IMP, bias 71.81 30.88 45.97 33.25 50.42 72.81 76.04 link link
MSDN, no bias 71.76 30.40 36.76 32.81 40.89 72.54 75.85 link link
MSDN, bias 71.48 30.22 34.49 32.58 38.71 72.45 75.62 link link
Neural Motif, bias 72.54 29.35 29.26 33.10 35.02 73.64 78.70 link link
GRCNN, bias 74.17 34.73 39.56 37.04 43.63 74.11 77.32 link link
RelDN 75.40 40.85 44.24 49.16 50.60 78.74 90.39 link link

Visual Genome

model sgdet@20 sgdet@50 sgdet@100 sgcls@20 sgcls@50 sgcls@100 predcls@20 predcls@50 predcls@100 model config
IMP, no bias 19.8 27.5 33.0 28.0 33.4 35.1 44.9 54.8 57.8 link link
IMP, bias 21.7 29.3 34.5 29.2 33.9 35.3 48.8 57.6 59.9 link link
MSDN, no bias 21.0 28.3 33.5 28.2 33.4 35.0 46.0 55.0 57.7 link link
MSDN, bias 22.4 30.0 35.3 29.7 34.4 35.9 51.2 59.6 61.6 link link
Neural Motif, no bias 21.0 28.6 33.8 29.2 34.1 35.5 51.0 60.2 62.3 link link
Neural Motif, bias 21.8 30.1 33.8 30.2 35.1 36.5 52.1 61.2 63.2 link link
GRCNN, no bias 20.5 27.4 35.7 27.1 31.5 32.9 42.5 50.7 53.3 link link
GRCNN, bias 22.9 30.1 34.8 30.5 34.9 36.2 52.1 59.9 61.8 link link
RelDN 24.0 32.4 37.8 31.9 35.7 36.6 54.0 60.9 62.5 link link