@miyane-ralo As you can seen in readme, my customize implementation of yolov5 relies on nb package. You can using some basic function such as build_conv_layer, build_act_layer to customize your model if you want, it's same style as mmdetection and detectron2, but expose more flexiblify and more simplier to use.
Also some out-of-box components such as various heads (PANet, SPP), some attention machinism were also supported for directly use.
And PR is very welcomed if you have your own components and it's really useful, you want integrating into nb, then just pull and request your mege!
@miyane-ralo As you can seen in readme, my customize implementation of yolov5 relies on nb package. You can using some basic function such as build_conv_layer, build_act_layer to customize your model if you want, it's same style as mmdetection and detectron2, but expose more flexiblify and more simplier to use.
Also some out-of-box components such as various heads (PANet, SPP), some attention machinism were also supported for directly use.
And PR is very welcomed if you have your own components and it's really useful, you want integrating into nb, then just pull and request your mege!