By Or Tslil, Tal Feiner
This work uses an implementation of [1], forked from the repository (https://github.com/balancap/SSD-Tensorflow). SSD is an unified framework for object detection with a single network. The intuition behaynd SSD is a multiple convolution operation with different shapes and sizes, each for different abstract (depth) of the network.
SSD architecture [1].
The output of such architechture is an array of possible object, each one with its predicted class, location in the image and abounding box. The loss function of SSD is a combination of a catagorical crossentropy and a mean square error (MSE) of the predicted location and bounding of each object.
Dependencies:
- install tensorflow (gpu is recomended)
- install opencv (pip install opencv-python --user)
- Clone the repository
- unzip
ssd_300_vgg.ckpt
file under themodel
folder.
[1] Liu W, Anguelov D, Erhan D, Szegedy C, Reed S, Fu CY, Berg AC. Ssd: Single shot multibox detector. InEuropean conference on computer vision 2016 Oct 8 (pp. 21-37). Springer, Cham.