YOLOv5 architecture from .pt files
Yolov5 is recently published by Glen Jocher from Ultralytics. Many controversial talks arise arguing between yolov4 and yolov5. However, the paper is still not published and according to the authors, we should expect by end of 2020.
Then the first question that comes in our mind is how this architecture looks like. I tried to create an overview of the architecture using netron visualizer for the neural network. The following results show after visualizing different pretrained checkpoint .pt files. Architectures or different pt files follow a similar structure except number of filters in Conv2D layers. Of course one needs to investigate further on the design of BottleneckCSP and SSP layers which is already explained in Yolov4. If you like it hit clap button and comment for questions or feedback.
