Home

Awesome

RetinaNet-mxnet

Adapted from SSD implemented by zhreshold, the results still need to be tuned. Currently we use the PASCAL VOC mAP metric which measures under IoU threshold 0.5, not the COCO AP metric.

Demo Results

image image image

Differences from SSD

Usage

For PASCAL VOC and more details, one can generally refer to SSD implemented by zhreshold.

Environment

Tested on Ubuntu 16.04, python3.5, mxnet 1.1.0

Numpy, cv2 and matplotlib are required.

mAP result

BackboneTraining dataVal dataStrategymAPNote
ResNet-50 512x512VOC07+12 trainvalVOC07 testOHEM76.0sgd, lr0.01
ResNet-50 512x512VOC07+12 trainvalVOC07 testFL75.4sgd, lr0.01
ResNet-50 512x512COCO2017 trainCOCO2017 valOHEM40.2sgd, lr0.01
ResNet-50 512x512COCO2017 trainCOCO2017 valFL40.9sgd, lr0.01

Baseline Faster RCNN

BackboneTraining dataVal datamAPNote
ResNet-50 600VOC07+12 trainvalVOC07 test74.8sgd, lr0.001
ResNet-50 600COCO2017 trainCOCO2017 val37.9sgd, lr0.003