search

RangiLyu / nanodet

NanoDet-Plusāš”Super fast and lightweight anchor-free object detection model. šŸ”„Only 980 KB(int8) / 1.8MB (fp16) and run 97FPS on cellphonešŸ”„
Apache License 2.0
5.71k stars 1.04k forks source link

input image size #543

Open mahilaMoghadami opened 10 months ago

mahilaMoghadami commented 10 months ago

hello does you code is train in various input image size? I mean that as it fully convolution, so maybe there is no need to resizing image before training? Am I right? if no what is best input image size for visdrone dataset? is there any paper for this code?

thank you