Models - MalondaClement/pipeline GitHub Wiki

Models Module

Models📦

List of all models available
  • Pytorch :
    • DeepLabV3_Resnet50;
    • DeepLabV3_Resnet101;
    • DeepLabV3_MobileNetV3;
    • FCN_Resnet50;
    • FCN_Resnet101.
  • External :
    • DenseASPP121;
    • DenseASPP161;
    • DenseASPP169;
    • DenseASPP201;
    • MobileNetDenseASPP;
    • UNet;

DenseASPP

class models.DenseASPP.DenseASPP(self, model_cfg, num_classes=19, output_stride=8)

[source]

Parameters
  • model_cfg - Configure model (configurations are in models.configs). Four configurations are available for this model: DenseASPP121, DenseASPP161, DenseASPP169 and DenseASPP201.
  • num_classes - Number of classes in the dataset
  • output_stride - Configure output size (can only be 8 or 16)
Returns
  • DenseASPP
Return type
  • torch.nn.Module

MobileNetDenseASPP

class models.MobileNetDenseASPP.DenseASPP(self, model_cfg, num_classes=19, output_stride=8)

[source]

Parameters
  • model_cfg - Configure model (configurations are in models.configs). Only one config is available for this model: MobileNetDenseASPP.
  • num_classes - Number of classes in the dataset
  • output_stride - Configure output size (can only be 8 or 16)
Returns
  • MobileNetDenseASPP
Return type
  • torch.nn.Module

UNet

class models.UNet.UNet(self, num_classes, batchnorm=False)

[source]

Parameters
  • num_classes - Number of classes in the dataset
  • batchnorm -
Returns
  • UNet
Return type
  • torch.nn.Module

Utils 🧰

[source]

Get Model

models.utils.get_model(args)

This function takes args object and uses args.model value to return model. Updates args.is_pytorch_model value and return model and args.

Parameters
  • args - Object with all arguments used during training.
Returns
  • (model, args)
Return type
  • torch.nn.Module, helpers.ARGS.ARGS
Example
from helpers.ARGS import ARGS
from datasets.tunnel import Tunnel

Dataset = Tunnel

args = ARGS("DenseASPP121", "dataset_path", len(Dataset.validClasses), labels_type="csv", batch_size=8, epochs=300)

model, args = get_model(args)

Get Optimizer (:warning: NOT IMPLEMENTED)

models.utils.get_optimizer(args)
Parameters
  • args - Object with all arguments used during training.