This is a re-implemented and re-trained version of tiny YOLO v2 object detection network trained with VOC2012 training dataset. Network weight pruning is applied to sparsify convolution layers (60% of network parameters are set to zeros).
| Metric | Value |
|---|---|
| Mean Average Precision (mAP) | 35.32% |
| Flops | 6.97Bn* |
| Source framework | Tensorflow** |
Average Precision metric described in: Mark Everingham et al. "The PASCAL Visual Object Classes (VOC) Challenge".
Tested on VOC 2012 validation dataset.
- name: "input" , shape: [1x3x416x416] - An input image in the format [BxCxHxW],
where:
- B - batch size
- C - number of channels
- H - image height
- W - image width. Expected color order is BGR.
- The net outputs a blob with the shape: [1, 21125], which can be reshaped to [5, 25, 13, 13],
where each number corresponds to [
num_anchors,cls_reg_obj_params,y_loc,x_loc], respectively.num_anchors: number of anchor boxes, each spatial location specified byy_locandx_lochas 5 anchors.cls_reg_obj_params: parameters for classification and regression. The values are made up of the followings:- Regression parameters (4)
- Objectness score (1)
- Class score (20)
y_locandx_loc: spatial location of each grid.
[*] Same as the original implementation.
[**] Other names and brands may be claimed as the property of others.