search

hendrycks / anomaly-seg

The Combined Anomalous Object Segmentation (CAOS) Benchmark
MIT License
154 stars 20 forks source link

BDD dataset training yaml #4

Closed YingdaXia closed 4 years ago

YingdaXia commented 4 years ago

Hello authors,

Thank you for your hard work! My question is that what's the training hyper parameters for BDD dataset? Is it the same as Street Hazards?

Thanks, Yingda

xksteven commented 4 years ago

We kept both hyperparameters the same on BDD and StreetHazards. We used the defaults from ADE20K.

To get improved performance using a batchsize larger than 2 helps out significantly. We also found that UPerNet50 performed better than the dilated ResNet50 + PPM_deepsup backbone although this effect was marginal.