MobileNetV1, MobileNetV2, VGG based SSD/SSD-lite implementation in Pytorch 1.0 / Pytorch 0.4. Out-of-box support for retraining on Open Images dataset. ONNX and Caffe2 support. Experiment Ideas like CoordConv.
When using the --resume feature of train_ssd.py, it appears as though the learning rate and other optimizations from the previous session are not retained. For example, if the learning rate scheduler has set learning rate to 0.00001 by the final epoch of a session, resuming from that particular epoch will begin training at the default learning rate.
By what means would one effectively use the --resume feature of train_ssd.py and resume training according to the previous session's learning rate scheduler?
When using the
--resume
feature oftrain_ssd.py
, it appears as though the learning rate and other optimizations from the previous session are not retained. For example, if the learning rate scheduler has set learning rate to 0.00001 by the final epoch of a session, resuming from that particular epoch will begin training at the default learning rate.By what means would one effectively use the
--resume
feature oftrain_ssd.py
and resume training according to the previous session's learning rate scheduler?