Object Detection

COCO

The evaluation was made on 5K validation set from COCO 2017
ModelFull Precision mAPQuantized mAPInput Resolution (HxWxC)Params (M)Flops (G)Pretrained ModelSource
centernet_resnet_v1_18^26.3724.57512x512x314.2315.73pretrainedlink
centernet_resnet_v1_50^31.7930.09512x512x330.0828.56pretrainedlink
ssd_mobiledet_dsp ^28.928.47320x320x37.072.83pretrainedlink
ssd_mobilenet_v1^23.1822.2300x300x36.791.25pretrainedlink
ssd_mobilenet_v1_hd17.6716.43720x1280x36.8112.26pretrainedlink
ssd_mobilenet_v224.2723.26300x300x34.460.76pretrainedlink
ssd_resnet_v1_1817.6817.28300x300x316.623.54pretrainedlink
tiny_yolov314.2212.84416x416x38.852.79pretrainedlink
tiny_yolov418.8916.39416x416x36.053.46pretrainedlink
yolov3_41636.0635.37416x416x368.7974.02pretrainedlink
yolov337.2136.96608x608x368.7979.17pretrainedlink
yolov442.440.6512x512x366.0547.36pretrainedlink
yolov5xs**32.7431.39512x512x37.915.8pretrainedlink
yolov5s**^34.3633.01640x640x37.919.06pretrainedlink
yolov5m**^41.739.12640x640x322.9527.33pretrainedlink

** Trained without the SPP block
^ networks available in TAPPAS

Visdrone

The model was evaluated on VisDrone2019-DET dataset.
ModelFull Precision mAPQuantized mAPInput Resolution (HxWxC)Params (M)Flops (G)Pretrained ModelSource
ssd_mobilenet_v1^2.3762.2300x300x35.641.15pretrained
link

^  networks available in TAPPAS