This is an action detector for the Smart Classroom scenario. It is based on the RMNet backbone that includes depth-wise convolutions to reduce the amount of computations for the 3x3 convolution block. The first SSD head from 1/16 scale feature map has four clustered prior boxes and outputs detected persons (two class detector). The second SSD-based head predicts actions of the detected persons. Possible actions: standing, writing, demonstrating.
Metric | Value |
---|---|
Detector AP (internal test set 2) | 77.11% |
Accuracy (internal test set 1) | 92.64% |
Pose coverage | Standing, writing, demonstrating |
Support of occluded pedestrians | YES |
Occlusion coverage | <50% |
Min pedestrian height | 80 pixels (on 1080p) |
GFlops | 7.140 |
MParams | 1.951 |
Source framework | Caffe* |
Average Precision (AP) is defined as an area under the precision/recall curve.
name: "input" , shape: [1x3x400x680] - An input image in the format [BxCxHxW], where:
Expected color order is BGR.
The net outputs four branches:
mbox_loc1/out/conv/flat
, shape: [b, num_priors*4] - Box coordinates in SSD formatmbox_main_conf/out/conv/flat/softmax/flat
, shape: [b, num_priors*2] - Detection confidencesmbox/priorbox
, shape: [1, 2, num_priors*4] - Prior boxes in SSD formatout/anchor1
, shape: [b, 3, h, w] - Action confidencesout/anchor2
, shape: [b, 3, h, w] - Action confidencesout/anchor3
, shape: [b, 3, h, w] - Action confidencesout/anchor4
, shape: [b, 3, h, w] - Action confidencesWhere:
[*] Other names and brands may be claimed as the property of others.