| # optimizer | |
| optimizer = dict(type='SGD', lr=0.02, momentum=0.9, weight_decay=0.0001) | |
| optimizer_config = dict(grad_clip=None) | |
| # learning policy | |
| lr_config = dict( | |
| policy='step', | |
| warmup='linear', | |
| warmup_iters=500, | |
| warmup_ratio=0.001, | |
| step=[8, 11]) | |
| runner = dict(type='EpochBasedRunner', max_epochs=12) | |