| import json |
|
|
| class Config(object): |
| def __init__(self, config_file): |
| with open(config_file) as f: |
| cfgs = json.load(f) |
| |
| |
| self.lr_backbone = cfgs["optimizer"]["lr_backbone"] |
| self.lr = cfgs["optimizer"]["lr"] |
|
|
| |
| self.epochs = cfgs["optimizer"]["epochs"] |
| self.warmup = cfgs["optimizer"]["warmup"] |
| self.warmup_epochs = cfgs["optimizer"]["warmup_epochs"] |
| self.lr_milestones = cfgs["optimizer"]["lr_milestones"] |
|
|
| self.start_epoch = cfgs["optimizer"]["start_epoch"] |
| self.weight_decay = cfgs["optimizer"]["weight_decay"] |
|
|
| |
| |
| self.backbone = cfgs["backbone"]["network"] |
| if self.backbone == 'resnet50' or self.backbone == 'resnet101': |
| self.dilation = True |
| elif self.backbone == 'resnet34' or self.backbone == 'resnet18': |
| self.dilation = False |
| else: |
| raise ValueError(f"{self.backbone} is not a supported backbone!") |
|
|
| self.position_embedding = cfgs["backbone"]["position_embedding"] |
| self.Frozen_BatchNorm2d = cfgs["backbone"]["Frozen_BatchNorm2d"] |
|
|
| |
| self.batch_size = cfgs["optimizer"]["batch_size"] |
| self.clip_max_norm = cfgs["optimizer"]["clip_max_norm"] |
|
|
| |
| self.SOS_token_id = cfgs["transformer"]["SOS_token_id"] |
| self.EOS_token_id = cfgs["transformer"]["EOS_token_id"] |
| self.PAD_token_id = cfgs["transformer"]["PAD_token_id"] |
|
|
| self.smooth = cfgs["transformer"]["smooth"] |
| self.dynamic_scale = cfgs["transformer"]["dynamic_scale"] |
|
|
| self.max_position_embeddings = cfgs["transformer"]["max_position_embeddings"] |
| self.vocab_size = cfgs["transformer"]["vocab_size"] |
|
|
|
|
| self.layer_norm_eps = cfgs["transformer"]["layer_norm_eps"] |
| self.dropout = cfgs["transformer"]["dropout"] |
|
|
| self.hidden_dim = cfgs["transformer"]["hidden_dim"] |
| self.enc_layers = cfgs["transformer"]["enc_layers"] |
| self.dec_layers = cfgs["transformer"]["dec_layers"] |
| self.dim_feedforward = cfgs["transformer"]["dim_feedforward"] |
| self.nheads = cfgs["transformer"]["nheads"] |
| self.pre_norm = cfgs["transformer"]["pre_norm"] |
|
|
| |
| self.imgsize = cfgs["dataset"]["imgsize"] |
