|
|
@@ -739,7 +739,7 @@ class RTCTrainer(object):
|
|
|
self.criterion = criterion
|
|
|
self.world_size = world_size
|
|
|
self.grad_accumulate = args.grad_accumulate
|
|
|
- self.clip_grad = 10
|
|
|
+ self.clip_grad = 35
|
|
|
self.heavy_eval = False
|
|
|
# weak augmentatino stage
|
|
|
self.second_stage = False
|
|
|
@@ -751,10 +751,8 @@ class RTCTrainer(object):
|
|
|
os.makedirs(self.path_to_save, exist_ok=True)
|
|
|
|
|
|
# ---------------------------- Hyperparameters refer to RTMDet ----------------------------
|
|
|
- # self.optimizer_dict = {'optimizer': 'adamw', 'momentum': None, 'weight_decay': 5e-2, 'lr0': 0.001}
|
|
|
- # self.ema_dict = {'ema_decay': 0.9998, 'ema_tau': 2000}
|
|
|
- self.optimizer_dict = {'optimizer': 'sgd', 'momentum': 0.9, 'weight_decay': 5e-4, 'lr0': 0.01}
|
|
|
- self.ema_dict = {'ema_decay': 0.9999, 'ema_tau': 2000}
|
|
|
+ self.optimizer_dict = {'optimizer': 'adamw', 'momentum': None, 'weight_decay': 5e-2, 'lr0': 0.001}
|
|
|
+ self.ema_dict = {'ema_decay': 0.9998, 'ema_tau': 2000}
|
|
|
self.lr_schedule_dict = {'scheduler': 'cosine', 'lrf': 0.05}
|
|
|
self.warmup_dict = {'warmup_momentum': 0.8, 'warmup_bias_lr': 0.1}
|
|
|
|
yjh0410