List your company
ResearchFormerly WizWand

Face Detection on FDDB

92Recall @ 1% FPPITiny-YOLO
83.1685.45587.7590.045Jul 29, 2017Dec 16, 2017May 6, 2018Sep 23, 2018Feb 11, 2019Jul 1, 2019Nov 19, 2019
Updated 8mo ago

Evaluation results

MethodLinks
Tiny-YOLOmodel size (MB)=63.002019.1192——
Tiny-YOLO#FLOPS (million)=24,407, Model size(KB)=62,5162019.1192——
Proposed (RSA+LRN)Scale-forecast=tiny ResNet-18, RSA=4-layer FCN, LRN=tiny ResNet-18, Opts. (VGA input)=95.67M to 1.5G2017.0791.92——
RPN_sStructure=single anchor, Opts. (VGA input)=1.72G, Backbone=ResNet-182017.0790.61——
DupNet-L+PACT#FLOPS (million)=95.7, Model size(KB)=45.4, clipping thresholds=PACT2019.1190.6——
Tinier-YOLO#FLOPS (million)=3,213, Model size(KB)=7,7072019.1190.2——
Tinier-YOLOmodel size (MB)=7.892019.1190——
IFQ-Tiny-YOLOmodel size (MB)=1.97, k=22019.1189——
DupNet-L#FLOPS (million)=95.7, Model size(KB)=45.42019.1188.4——
DupNet+PACT#FLOPS (million)=62.6, Model size(KB)=36.9, clipping thresholds=PACT2019.1188——
RPN_mStructure=multi anchors, Opts. (VGA input)=1.31G, Backbone=ResNet-182017.0786.89——
DupNet#FLOPS (million)=62.6, Model size(KB)=36.92019.1185.9——
IFQ-Tinier-YOLOmodel size (MB)=0.25, k=22019.1184——
IFQ-Tinier-YOLO#FLOPS (million)=107.9, Model size(KB)=240.92019.1183.5——
ASFD2021.05——99.11
Condensation-NetAlpha=2, Pooling Type=Max Pooling, Network Precision=Quantized2021.04—91.82—
Condensation-NetAlpha=2, Pooling Type=Max Pooling, Network Precision=Full-Precision2021.04—93.86—
Condensation-NetAlpha=2, Pooling Type=Average Pooling, Network Precision=Quantized2021.04—91.13—
Condensation-NetAlpha=2, Pooling Type=Average Pooling, Network Precision=Full-Precision2021.04—93.69—
Condensation-NetAlpha=4, Pooling Type=Max Pooling, Network Precision=Quantized2021.04—90.25—
Condensation-NetAlpha=4, Pooling Type=Max Pooling, Network Precision=Full-Precision2021.04—93.61—
Condensation-NetAlpha=4, Pooling Type=Average Pooling, Network Precision=Quantized2021.04—90.74—
Condensation-NetAlpha=4, Pooling Type=Average Pooling, Network Precision=Full-Precision2021.04—93.56—
FaceBoxes2021.05——95.98
PyramidBox2021.05——98.69
RefineFace2021.05——99.11
Tiny-YOLOv2Network Precision=Quantized2021.04—89.87—
Tiny-YOLOv2Network Precision=Full-Precision2021.04—92.28—
YOLOv5lvariant=large2021.05——98.67
YOLOv5l6variant=large-62021.05——98.8
YOLOv5mvariant=medium2021.05——98.49
YOLOv5svariant=small2021.05——98.43