search

YuHengsss / YOLOV

This repo is an implementation of PyTorch version YOLOV Series
Apache License 2.0
278 stars 39 forks source link

YOLOX outperforms YOLOV on different dataset using "vid" optimization settings #52

Closed akanuasiegbu closed 1 year ago

akanuasiegbu commented 1 year ago

Hi, do you have any tips on how to improve YOLOV performance? I'm training YOLOV on a different dataset and numerically found that YOLOX produced better results. Maybe it's hyperparameter tuning specific to the dataset? To be clear I trained YOLOX on this particular dataset using the hyperparameter settings in "yoloxs_vid.py". Then used the trained YOLOX model to initialize the YOLOV backbone and trained YOLOV using the hyperparameter setting in "yolovs_vid.py". Note that I am using only local frames and I am not using any global frames.

YuHengsss commented 1 year ago

Hi there, you could expand the stride (defualt=1) of sampling local frames which could boost detection accuracy.