search

hnuzhy / SSDA-YOLO

Codes for my paper "SSDA-YOLO: Semi-supervised Domain Adaptive YOLO for Cross-Domain Object Detection"
152 stars 20 forks source link

How to trian on my custom dataset? #4

Open liuhaolinwen opened 1 year ago

liuhaolinwen commented 1 year ago

Hello author, i want to trian on my custom dataset, what should i do?

hnuzhy commented 1 year ago

Hello author, i want to trian on my custom dataset, what should i do?

Training a custom dataset may be splitted into following four steps: 1) Preparing your source domain train-set&val-set with object detection GT labels, and target domain train-set (large-scale without labels) & val-set (small-scale with labels) for performance validation. 2) Training the image style transfer CUT model offline, which includes both source→target and target→source for generating corresponding fake images. More deatils can be found in its official code link https://github.com/taesungp/contrastive-unpaired-translation 3) Convert all images and labels data into formats that our project needs. Please refer the scripts https://github.com/hnuzhy/SSDA-YOLO/tree/master/data/formats/ 4) Training on your custom dataset following our instructions in README.md