The model is composed of 161 layers.
Most of them are Conv2D, there are also 3 … See more

The yolo weight have been retreived from https://githu…
The file contain the kernel weights but also the biases and the Batch Normalisation parameter… See more

During the pre-processing the 80 labels and the image to predict are loaded.
The labels are in the file coco_classes.txt.
The image is resized in the Yolo format 608*608 us… See more

The model is run with the resized image as input with a shape=(1,608,608,3).
The model provides 3 output layers 139, 150 and 161 with the shapes respectively (1, 76, 76, … See more

As explained before, the 3 final Yolo layers are computed outside the TF model by the python function decode_netout.
The steps of this function are the following:
•apply the s… See more