Lightning Modules

class datamint.lightning.trainers.lightning_modules.ClassificationModule(model_name, num_classes, loss_fn, metrics_factories, class_names, image_size, lr=0.0001, pretrained=True, transform=None)

Bases: DatamintLightningModule

Parameters:

  • model_name (str)

  • num_classes (int)

  • loss_fn (Module)

  • metrics_factories (dict[str, Callable[[], Any]])

  • class_names (list[tuple[str, str]])

  • image_size (tuple[int, int] | None)

  • lr (float)

  • pretrained (bool)

  • transform (BasicTransform | BaseCompose | None)

configure_optimizers()

Choose what optimizers and learning-rate schedulers to use in your optimization. Normally you’d need one. But in the case of GANs or similar you might have multiple. Optimization with multiple optimizers only works in the manual optimization mode.

Returns:

Any of these 6 options.

  • Single optimizer.

  • List or Tuple of optimizers.

  • Two lists - The first list has multiple optimizers, and the second has multiple LR schedulers (or multiple lr_scheduler_config).

  • Dictionary, with an "optimizer" key, and (optionally) a "lr_scheduler" key whose value is a single LR scheduler or lr_scheduler_config.

  • None - Fit will run without any optimizer.

The lr_scheduler_config is a dictionary which contains the scheduler and its associated configuration. The default configuration is shown below.

lr_scheduler_config = {
    # REQUIRED: The scheduler instance
    "scheduler": lr_scheduler,
    # The unit of the scheduler's step size, could also be 'step'.
    # 'epoch' updates the scheduler on epoch end whereas 'step'
    # updates it after a optimizer update.
    "interval": "epoch",
    # How many epochs/steps should pass between calls to
    # `scheduler.step()`. 1 corresponds to updating the learning
    # rate after every epoch/step.
    "frequency": 1,
    # Metric to monitor for schedulers like `ReduceLROnPlateau`
    "monitor": "val_loss",
    # If set to `True`, will enforce that the value specified 'monitor'
    # is available when the scheduler is updated, thus stopping
    # training if not found. If set to `False`, it will only produce a warning
    "strict": True,
    # If using the `LearningRateMonitor` callback to monitor the
    # learning rate progress, this keyword can be used to specify
    # a custom logged name
    "name": None,
}

When there are schedulers in which the .step() method is conditioned on a value, such as the torch.optim.lr_scheduler.ReduceLROnPlateau scheduler, Lightning requires that the lr_scheduler_config contains the keyword "monitor" set to the metric name that the scheduler should be conditioned on.

Metrics can be made available to monitor by simply logging it using self.log('metric_to_track', metric_val) in your LightningModule.

Note

Some things to know:

  • Lightning calls .backward() and .step() automatically in case of automatic optimization.

  • If a learning rate scheduler is specified in configure_optimizers() with key "interval" (default “epoch”) in the scheduler configuration, Lightning will call the scheduler’s .step() method automatically in case of automatic optimization.

  • If you use 16-bit precision (precision=16), Lightning will automatically handle the optimizer.

  • If you use torch.optim.LBFGS, Lightning handles the closure function automatically for you.

  • If you use multiple optimizers, you will have to switch to ‘manual optimization’ mode and step them yourself.

  • If you need to control how often the optimizer steps, override the optimizer_step() hook.

forward(x)

Same as torch.nn.Module.forward().

Parameters:

  • *args – Whatever you decide to pass into the forward method.

  • **kwargs – Keyword arguments are also possible.

  • x (Tensor)

Return type:

Tensor

Returns:

Your model’s output

on_test_epoch_end()

Called in the test loop at the very end of the epoch.

Return type:

None

on_train_epoch_end()

Called in the training loop at the very end of the epoch.

To access all batch outputs at the end of the epoch, you can cache step outputs as an attribute of the LightningModule and access them in this hook:

class MyLightningModule(L.LightningModule):
    def __init__(self):
        super().__init__()
        self.training_step_outputs = []

    def training_step(self):
        loss = ...
        self.training_step_outputs.append(loss)
        return loss

    def on_train_epoch_end(self):
        # do something with all training_step outputs, for example:
        epoch_mean = torch.stack(self.training_step_outputs).mean()
        self.log("training_epoch_mean", epoch_mean)
        # free up the memory
        self.training_step_outputs.clear()
Return type:

None

on_validation_epoch_end()

Called in the validation loop at the very end of the epoch.

Return type:

None

predict_default(model_input, compute_uncertainty=False, **kwargs)

Run classification inference, returning ImageClassification per resource.

Parameters:

  • compute_uncertainty (bool) – If True, also compute a predictive-entropy uncertainty score (see datamint.utils.uncertainty) and attach it as uncertainty on each returned annotation.

  • kwargs (Any)

task_type = 'image_classification'

Generic image classification module backed by timm.

Parameters:

  • model_name – timm model name (e.g. 'resnet34', 'efficientnet_b0').

  • num_classes – Number of output classes.

  • loss_fn – Loss module.

  • metrics_factories – {name: callable} – see SegmentationModule.

  • lr – Learning rate for AdamW.

  • pretrained – Use pretrained weights.

  • image_size – Optional inference resize target (H, W). When omitted, predictions keep the original image size.

test_step(batch, batch_idx)

Operates on a single batch of data from the test set. In this step you’d normally generate examples or calculate anything of interest such as accuracy.

Parameters:

  • batch (dict) – The output of your data iterable, normally a DataLoader.

  • batch_idx (int) – The index of this batch.

  • dataloader_idx – The index of the dataloader that produced this batch. (only if multiple dataloaders used)

Return type:

Tensor

Returns:

  • Tensor - The loss tensor

  • dict - A dictionary. Can include any keys, but must include the key 'loss'.

  • None - Skip to the next batch.

# if you have one test dataloader:
def test_step(self, batch, batch_idx): ...


# if you have multiple test dataloaders:
def test_step(self, batch, batch_idx, dataloader_idx=0): ...

Examples:

# CASE 1: A single test dataset
def test_step(self, batch, batch_idx):
    x, y = batch

    # implement your own
    out = self(x)
    loss = self.loss(out, y)

    # log 6 example images
    # or generated text... or whatever
    sample_imgs = x[:6]
    grid = torchvision.utils.make_grid(sample_imgs)
    self.logger.experiment.add_image('example_images', grid, 0)

    # calculate acc
    labels_hat = torch.argmax(out, dim=1)
    test_acc = torch.sum(y == labels_hat).item() / (len(y) * 1.0)

    # log the outputs!
    self.log_dict({'test_loss': loss, 'test_acc': test_acc})

If you pass in multiple test dataloaders, test_step() will have an additional argument. We recommend setting the default value of 0 so that you can quickly switch between single and multiple dataloaders.

# CASE 2: multiple test dataloaders
def test_step(self, batch, batch_idx, dataloader_idx=0):
    # dataloader_idx tells you which dataset this is.
    x, y = batch

    # implement your own
    out = self(x)

    if dataloader_idx == 0:
        loss = self.loss0(out, y)
    else:
        loss = self.loss1(out, y)

    # calculate acc
    labels_hat = torch.argmax(out, dim=1)
    acc = torch.sum(y == labels_hat).item() / (len(y) * 1.0)

    # log the outputs separately for each dataloader
    self.log_dict({f"test_loss_{dataloader_idx}": loss, f"test_acc_{dataloader_idx}": acc})

Note

If you don’t need to test you don’t need to implement this method.

Note

When the test_step() is called, the model has been put in eval mode and PyTorch gradients have been disabled. At the end of the test epoch, the model goes back to training mode and gradients are enabled.

training_step(batch, batch_idx)

Here you compute and return the training loss and some additional metrics for e.g. the progress bar or logger.

Parameters:

  • batch (dict) – The output of your data iterable, normally a DataLoader.

  • batch_idx (int) – The index of this batch.

  • dataloader_idx – The index of the dataloader that produced this batch. (only if multiple dataloaders used)

Return type:

Tensor

Returns:

  • Tensor - The loss tensor

  • dict - A dictionary which can include any keys, but must include the key 'loss' in the case of automatic optimization.

  • None - In automatic optimization, this will skip to the next batch (but is not supported for multi-GPU, TPU, or DeepSpeed). For manual optimization, this has no special meaning, as returning the loss is not required.

In this step you’d normally do the forward pass and calculate the loss for a batch. You can also do fancier things like multiple forward passes or something model specific.

Example:

def training_step(self, batch, batch_idx):
    x, y, z = batch
    out = self.encoder(x)
    loss = self.loss(out, x)
    return loss

To use multiple optimizers, you can switch to ‘manual optimization’ and control their stepping:

def __init__(self):
    super().__init__()
    self.automatic_optimization = False


# Multiple optimizers (e.g.: GANs)
def training_step(self, batch, batch_idx):
    opt1, opt2 = self.optimizers()

    # do training_step with encoder
    ...
    opt1.step()
    # do training_step with decoder
    ...
    opt2.step()

Note

When accumulate_grad_batches > 1, the loss returned here will be automatically normalized by accumulate_grad_batches internally.

validation_step(batch, batch_idx)

Operates on a single batch of data from the validation set. In this step you’d might generate examples or calculate anything of interest like accuracy.

Parameters:

  • batch (dict) – The output of your data iterable, normally a DataLoader.

  • batch_idx (int) – The index of this batch.

  • dataloader_idx – The index of the dataloader that produced this batch. (only if multiple dataloaders used)

Return type:

Tensor

Returns:

  • Tensor - The loss tensor

  • dict - A dictionary. Can include any keys, but must include the key 'loss'.

  • None - Skip to the next batch.

# if you have one val dataloader:
def validation_step(self, batch, batch_idx): ...


# if you have multiple val dataloaders:
def validation_step(self, batch, batch_idx, dataloader_idx=0): ...

Examples:

# CASE 1: A single validation dataset
def validation_step(self, batch, batch_idx):
    x, y = batch

    # implement your own
    out = self(x)
    loss = self.loss(out, y)

    # log 6 example images
    # or generated text... or whatever
    sample_imgs = x[:6]
    grid = torchvision.utils.make_grid(sample_imgs)
    self.logger.experiment.add_image('example_images', grid, 0)

    # calculate acc
    labels_hat = torch.argmax(out, dim=1)
    val_acc = torch.sum(y == labels_hat).item() / (len(y) * 1.0)

    # log the outputs!
    self.log_dict({'val_loss': loss, 'val_acc': val_acc})

If you pass in multiple val dataloaders, validation_step() will have an additional argument. We recommend setting the default value of 0 so that you can quickly switch between single and multiple dataloaders.

# CASE 2: multiple validation dataloaders
def validation_step(self, batch, batch_idx, dataloader_idx=0):
    # dataloader_idx tells you which dataset this is.
    x, y = batch

    # implement your own
    out = self(x)

    if dataloader_idx == 0:
        loss = self.loss0(out, y)
    else:
        loss = self.loss1(out, y)

    # calculate acc
    labels_hat = torch.argmax(out, dim=1)
    acc = torch.sum(y == labels_hat).item() / (len(y) * 1.0)

    # log the outputs separately for each dataloader
    self.log_dict({f"val_loss_{dataloader_idx}": loss, f"val_acc_{dataloader_idx}": acc})

Note

If you don’t need to validate you don’t need to implement this method.

Note

When the validation_step() is called, the model has been put in eval mode and PyTorch gradients have been disabled. At the end of validation, the model goes back to training mode and gradients are enabled.

class datamint.lightning.trainers.lightning_modules.DatamintLightningModule(settings=None, transform=None)

Bases: LightningModule, BaseDatamintModel

A LightningModule that is also a BaseDatamintModel.

Built-in trainers use this as the base for their default models so that the trained module can be logged once with datamint_flavor — no separate adapter step is required.

Inherits from both LightningModule and BaseDatamintModel.

Parameters:

  • settings (ModelSettings | None)

  • transform (BasicTransform | BaseCompose | None)

enable_sample_logging(enabled=True)

Enable or disable per-sample metric accumulation.

Parameters:

enabled (bool)

Return type:

None

load_context(context)

Move weights to the configured device and set eval mode on MLflow load.

Parameters:

context (PythonModelContext)

Return type:

None

mlflow_model_id: str | None
on_test_epoch_end()

Called in the test loop at the very end of the epoch.

Return type:

None

on_test_start()

Called at the beginning of testing.

Return type:

None

class datamint.lightning.trainers.lightning_modules.DeepLabV3PlusModule(in_channels, num_classes, loss_fn=None, metrics_factories={}, class_names=None, image_size=None, lr=0.0001, encoder_name='resnet34', encoder_weights='imagenet', decoder_atrous_rates=(12, 24, 36), transform=None)

Bases: SMPSegmentationModule

Segmentation module using the DeepLab v3+ architecture from segmentation_models_pytorch.

Parameters:

  • in_channels (int) – Number of input image channels.

  • num_classes (int) – Number of segmentation classes excluding background.

  • loss_fn (Module | None) – Loss module.

  • metrics_factories (dict[str, Callable[[], Any]]) – {name: callable} where each callable returns a fresh metric.

  • class_names (list[str] | None) – Human-readable label for each class.

  • image_size (tuple[int, int] | None) – (height, width) used during inference.

  • lr (float) – Learning rate for AdamW.

  • encoder_name (str) – SMP encoder backbone (e.g. 'resnet34', 'efficientnet-b4').

  • encoder_weights (str | None) – Pre-trained weights to load. 'imagenet' by default.

  • decoder_atrous_rates (tuple[int, int, int]) – Dilation rates for the ASPP module. Controls the multi-scale receptive field that is DeepLab v3+’s core architectural feature. SMP default is (12, 24, 36).

  • transform (BasicTransform | BaseCompose | None) – Albumentations transform applied during inference.

forward(x)

Same as torch.nn.Module.forward().

Parameters:

  • *args – Whatever you decide to pass into the forward method.

  • **kwargs – Keyword arguments are also possible.

  • x (Tensor)

Return type:

Tensor

Returns:

Your model’s output

class datamint.lightning.trainers.lightning_modules.SMPSegmentationModule(loss_fn=None, metrics_factories={}, class_names=None, transform=None, lr=0.0001)

Bases: SegmentationModule

Base segmentation module for architectures from segmentation_models_pytorch.

Handles SMP-specific construction parameters shared across all SMP architectures. Subclasses implement _build_model() to return the concrete SMP model.

Parameters:

  • loss_fn (Module | None)

  • metrics_factories (dict[str, Callable[[], Any]])

  • class_names (list[str] | None)

  • transform (BasicTransform | BaseCompose | None)

  • lr (float)

class datamint.lightning.trainers.lightning_modules.SegmentationModule(loss_fn=None, metrics_factories={}, class_names=None, transform=None, lr=0.0001)

Bases: DatamintLightningModule

Parameters:

  • loss_fn (Module | None)

  • metrics_factories (dict[str, Callable[[], Any]])

  • class_names (list[str] | None)

  • transform (BasicTransform | BaseCompose | None)

  • lr (float)

configure_optimizers()

Choose what optimizers and learning-rate schedulers to use in your optimization. Normally you’d need one. But in the case of GANs or similar you might have multiple. Optimization with multiple optimizers only works in the manual optimization mode.

Returns:

Any of these 6 options.

  • Single optimizer.

  • List or Tuple of optimizers.

  • Two lists - The first list has multiple optimizers, and the second has multiple LR schedulers (or multiple lr_scheduler_config).

  • Dictionary, with an "optimizer" key, and (optionally) a "lr_scheduler" key whose value is a single LR scheduler or lr_scheduler_config.

  • None - Fit will run without any optimizer.

The lr_scheduler_config is a dictionary which contains the scheduler and its associated configuration. The default configuration is shown below.

lr_scheduler_config = {
    # REQUIRED: The scheduler instance
    "scheduler": lr_scheduler,
    # The unit of the scheduler's step size, could also be 'step'.
    # 'epoch' updates the scheduler on epoch end whereas 'step'
    # updates it after a optimizer update.
    "interval": "epoch",
    # How many epochs/steps should pass between calls to
    # `scheduler.step()`. 1 corresponds to updating the learning
    # rate after every epoch/step.
    "frequency": 1,
    # Metric to monitor for schedulers like `ReduceLROnPlateau`
    "monitor": "val_loss",
    # If set to `True`, will enforce that the value specified 'monitor'
    # is available when the scheduler is updated, thus stopping
    # training if not found. If set to `False`, it will only produce a warning
    "strict": True,
    # If using the `LearningRateMonitor` callback to monitor the
    # learning rate progress, this keyword can be used to specify
    # a custom logged name
    "name": None,
}

When there are schedulers in which the .step() method is conditioned on a value, such as the torch.optim.lr_scheduler.ReduceLROnPlateau scheduler, Lightning requires that the lr_scheduler_config contains the keyword "monitor" set to the metric name that the scheduler should be conditioned on.

Metrics can be made available to monitor by simply logging it using self.log('metric_to_track', metric_val) in your LightningModule.

Note

Some things to know:

  • Lightning calls .backward() and .step() automatically in case of automatic optimization.

  • If a learning rate scheduler is specified in configure_optimizers() with key "interval" (default “epoch”) in the scheduler configuration, Lightning will call the scheduler’s .step() method automatically in case of automatic optimization.

  • If you use 16-bit precision (precision=16), Lightning will automatically handle the optimizer.

  • If you use torch.optim.LBFGS, Lightning handles the closure function automatically for you.

  • If you use multiple optimizers, you will have to switch to ‘manual optimization’ mode and step them yourself.

  • If you need to control how often the optimizer steps, override the optimizer_step() hook.

abstractmethod forward(x)

Same as torch.nn.Module.forward().

Parameters:

  • *args – Whatever you decide to pass into the forward method.

  • **kwargs – Keyword arguments are also possible.

  • x (Tensor)

Return type:

Tensor

Returns:

Your model’s output

on_test_epoch_end()

Called in the test loop at the very end of the epoch.

Return type:

None

on_train_epoch_end()

Called in the training loop at the very end of the epoch.

To access all batch outputs at the end of the epoch, you can cache step outputs as an attribute of the LightningModule and access them in this hook:

class MyLightningModule(L.LightningModule):
    def __init__(self):
        super().__init__()
        self.training_step_outputs = []

    def training_step(self):
        loss = ...
        self.training_step_outputs.append(loss)
        return loss

    def on_train_epoch_end(self):
        # do something with all training_step outputs, for example:
        epoch_mean = torch.stack(self.training_step_outputs).mean()
        self.log("training_epoch_mean", epoch_mean)
        # free up the memory
        self.training_step_outputs.clear()
Return type:

None

on_validation_epoch_end()

Called in the validation loop at the very end of the epoch.

Return type:

None

predict_image(model_input, compute_uncertainty=False, **kwargs)

Run segmentation inference, returning ImageSegmentation per resource.

Parameters:

  • compute_uncertainty (bool) – If True, also compute a predictive-entropy uncertainty score per class (see datamint.utils.uncertainty, restricted to the pixels predicted as that class) and attach it as uncertainty on each returned annotation.

  • kwargs (Any)

task_type = 'image_segmentation'

Base segmentation module for semantic segmentation tasks.

Subclasses must implement _build_model() to return the model.

Parameters:

  • in_channels – Number of input channels.

  • num_classes – Number of segmentation classes excluding background.

  • loss_fn – Loss module.

  • metrics_factories – {name: callable} where each callable returns a fresh torchmetrics.Metric. One instance is created per stage (train / val / test).

  • class_names – Human-readable label for each class.

  • image_size – (height, width) used during inference.

  • lr – Learning rate for AdamW.

test_step(batch, batch_idx)

Operates on a single batch of data from the test set. In this step you’d normally generate examples or calculate anything of interest such as accuracy.

Parameters:

  • batch (dict) – The output of your data iterable, normally a DataLoader.

  • batch_idx (int) – The index of this batch.

  • dataloader_idx – The index of the dataloader that produced this batch. (only if multiple dataloaders used)

Return type:

Tensor

Returns:

  • Tensor - The loss tensor

  • dict - A dictionary. Can include any keys, but must include the key 'loss'.

  • None - Skip to the next batch.

# if you have one test dataloader:
def test_step(self, batch, batch_idx): ...


# if you have multiple test dataloaders:
def test_step(self, batch, batch_idx, dataloader_idx=0): ...

Examples:

# CASE 1: A single test dataset
def test_step(self, batch, batch_idx):
    x, y = batch

    # implement your own
    out = self(x)
    loss = self.loss(out, y)

    # log 6 example images
    # or generated text... or whatever
    sample_imgs = x[:6]
    grid = torchvision.utils.make_grid(sample_imgs)
    self.logger.experiment.add_image('example_images', grid, 0)

    # calculate acc
    labels_hat = torch.argmax(out, dim=1)
    test_acc = torch.sum(y == labels_hat).item() / (len(y) * 1.0)

    # log the outputs!
    self.log_dict({'test_loss': loss, 'test_acc': test_acc})

If you pass in multiple test dataloaders, test_step() will have an additional argument. We recommend setting the default value of 0 so that you can quickly switch between single and multiple dataloaders.

# CASE 2: multiple test dataloaders
def test_step(self, batch, batch_idx, dataloader_idx=0):
    # dataloader_idx tells you which dataset this is.
    x, y = batch

    # implement your own
    out = self(x)

    if dataloader_idx == 0:
        loss = self.loss0(out, y)
    else:
        loss = self.loss1(out, y)

    # calculate acc
    labels_hat = torch.argmax(out, dim=1)
    acc = torch.sum(y == labels_hat).item() / (len(y) * 1.0)

    # log the outputs separately for each dataloader
    self.log_dict({f"test_loss_{dataloader_idx}": loss, f"test_acc_{dataloader_idx}": acc})

Note

If you don’t need to test you don’t need to implement this method.

Note

When the test_step() is called, the model has been put in eval mode and PyTorch gradients have been disabled. At the end of the test epoch, the model goes back to training mode and gradients are enabled.

training_step(batch, batch_idx)

Here you compute and return the training loss and some additional metrics for e.g. the progress bar or logger.

Parameters:

  • batch (dict) – The output of your data iterable, normally a DataLoader.

  • batch_idx (int) – The index of this batch.

  • dataloader_idx – The index of the dataloader that produced this batch. (only if multiple dataloaders used)

Return type:

Tensor

Returns:

  • Tensor - The loss tensor

  • dict - A dictionary which can include any keys, but must include the key 'loss' in the case of automatic optimization.

  • None - In automatic optimization, this will skip to the next batch (but is not supported for multi-GPU, TPU, or DeepSpeed). For manual optimization, this has no special meaning, as returning the loss is not required.

In this step you’d normally do the forward pass and calculate the loss for a batch. You can also do fancier things like multiple forward passes or something model specific.

Example:

def training_step(self, batch, batch_idx):
    x, y, z = batch
    out = self.encoder(x)
    loss = self.loss(out, x)
    return loss

To use multiple optimizers, you can switch to ‘manual optimization’ and control their stepping:

def __init__(self):
    super().__init__()
    self.automatic_optimization = False


# Multiple optimizers (e.g.: GANs)
def training_step(self, batch, batch_idx):
    opt1, opt2 = self.optimizers()

    # do training_step with encoder
    ...
    opt1.step()
    # do training_step with decoder
    ...
    opt2.step()

Note

When accumulate_grad_batches > 1, the loss returned here will be automatically normalized by accumulate_grad_batches internally.

validation_step(batch, batch_idx)

Operates on a single batch of data from the validation set. In this step you’d might generate examples or calculate anything of interest like accuracy.

Parameters:

  • batch (dict) – The output of your data iterable, normally a DataLoader.

  • batch_idx (int) – The index of this batch.

  • dataloader_idx – The index of the dataloader that produced this batch. (only if multiple dataloaders used)

Return type:

Tensor

Returns:

  • Tensor - The loss tensor

  • dict - A dictionary. Can include any keys, but must include the key 'loss'.

  • None - Skip to the next batch.

# if you have one val dataloader:
def validation_step(self, batch, batch_idx): ...


# if you have multiple val dataloaders:
def validation_step(self, batch, batch_idx, dataloader_idx=0): ...

Examples:

# CASE 1: A single validation dataset
def validation_step(self, batch, batch_idx):
    x, y = batch

    # implement your own
    out = self(x)
    loss = self.loss(out, y)

    # log 6 example images
    # or generated text... or whatever
    sample_imgs = x[:6]
    grid = torchvision.utils.make_grid(sample_imgs)
    self.logger.experiment.add_image('example_images', grid, 0)

    # calculate acc
    labels_hat = torch.argmax(out, dim=1)
    val_acc = torch.sum(y == labels_hat).item() / (len(y) * 1.0)

    # log the outputs!
    self.log_dict({'val_loss': loss, 'val_acc': val_acc})

If you pass in multiple val dataloaders, validation_step() will have an additional argument. We recommend setting the default value of 0 so that you can quickly switch between single and multiple dataloaders.

# CASE 2: multiple validation dataloaders
def validation_step(self, batch, batch_idx, dataloader_idx=0):
    # dataloader_idx tells you which dataset this is.
    x, y = batch

    # implement your own
    out = self(x)

    if dataloader_idx == 0:
        loss = self.loss0(out, y)
    else:
        loss = self.loss1(out, y)

    # calculate acc
    labels_hat = torch.argmax(out, dim=1)
    acc = torch.sum(y == labels_hat).item() / (len(y) * 1.0)

    # log the outputs separately for each dataloader
    self.log_dict({f"val_loss_{dataloader_idx}": loss, f"val_acc_{dataloader_idx}": acc})

Note

If you don’t need to validate you don’t need to implement this method.

Note

When the validation_step() is called, the model has been put in eval mode and PyTorch gradients have been disabled. At the end of validation, the model goes back to training mode and gradients are enabled.

class datamint.lightning.trainers.lightning_modules.TransUNetModule(in_channels, num_classes, loss_fn=None, metrics_factories={}, class_names=None, image_size=None, lr=0.0001, variant='R50-ViT-B_16', pretrained=True, transform=None)

Bases: SegmentationModule

Segmentation module using the TransUNet architecture.

Uses a timm vit_*_r50_s*_224 hybrid backbone (ResNet-50 CNN stages + ViT transformer) combined with a four-block Cascaded UPsampler decoder.

Parameters:

  • in_channels (int) – Number of input image channels.

  • num_classes (int) – Number of segmentation classes excluding background.

  • loss_fn (Module | None) – Loss module.

  • metrics_factories (dict[str, Callable[[], Any]]) – {name: callable} where each callable returns a fresh metric.

  • class_names (list[str] | None) – Human-readable label for each class.

  • image_size (tuple[int, int] | None) – (height, width) used during inference (must be 224×224).

  • lr (float) – Learning rate for AdamW.

  • variant (str) – TransUNet variant name. One of 'R50-ViT-B_16' (default) or 'R50-ViT-L_16'.

  • pretrained (bool) – Load ImageNet-21k pre-trained ViT weights via timm.

  • transform (BasicTransform | BaseCompose | None) – Albumentations transform applied during inference.

forward(x)

Same as torch.nn.Module.forward().

Parameters:

  • *args – Whatever you decide to pass into the forward method.

  • **kwargs – Keyword arguments are also possible.

  • x (Tensor)

Return type:

Tensor

Returns:

Your model’s output

class datamint.lightning.trainers.lightning_modules.UNETRPPModule(in_channels, num_classes, img_size, loss_fn=None, metrics_factories={}, class_names=None, lr=0.0001, feature_size=16, num_heads=4, depths=None, patch_crop_size=(128, 128, 128), sw_overlap=0.5, transform=None)

Bases: SegmentationModule

Segmentation module wrapping UNETR++ for 3-D volumetric inputs.

Differences from 2-D modules: - training_step: random 3-D patch crop before forward pass. - validation_step / test_step: sliding-window inference over

full volumes with Gaussian importance weighting.

Parameters:

  • in_channels (int) – Number of input image channels (e.g. 1 for CT, 4 for MRI).

  • num_classes (int) – Number of segmentation classes excluding background.

  • img_size (tuple[int, int, int]) – Spatial size (D, H, W) of each training patch. Must equal patch_crop_size. Fixed due to positional embeddings.

  • feature_size (int) – Base channel width. Channel dims at each encoder stage are [2F, 4F, 8F, 16F] where F = feature_size. Default 16 matches the Synapse/ACDC configs from the paper.

  • num_heads (int) – Number of attention heads in EPA.

  • depths (list[int] | None) – Number of transformer blocks per encoder stage (list of 4).

  • patch_crop_size (tuple[int, int, int]) – (D, H, W) patch randomly cropped during training. Must match img_size.

  • sw_overlap (float) – Overlap ratio for sliding-window inference (0–1).

  • loss_fn (Module | None) – Loss module.

  • metrics_factories (dict[str, Callable[[], Any]]) – {name: callable} where each callable returns a fresh torchmetrics.Metric.

  • class_names (list[str] | None) – Human-readable label for each class.

  • lr (float) – Learning rate for AdamW.

  • transform (BasicTransform | BaseCompose | None)

forward(x)

Same as torch.nn.Module.forward().

Parameters:

  • *args – Whatever you decide to pass into the forward method.

  • **kwargs – Keyword arguments are also possible.

  • x (Tensor)

Return type:

Tensor

Returns:

Your model’s output

predict_volume(model_input, compute_uncertainty=False, uncertainty_top_fraction=0.2, **kwargs)

Run sliding-window segmentation inference, returning VolumeSegmentation per resource.

Parameters:

  • compute_uncertainty (bool) – If True, also compute a predictive-entropy uncertainty score per class (see datamint.utils.uncertainty), pooled from per-slice scores to one score per volume, and attach it as uncertainty on each returned annotation.

  • uncertainty_top_fraction (float) – Fraction of the volume’s most-uncertain slices averaged into the per-volume score. Only used when compute_uncertainty=True.

  • kwargs (Any)

test_step(batch, batch_idx)

Operates on a single batch of data from the test set. In this step you’d normally generate examples or calculate anything of interest such as accuracy.

Parameters:

  • batch (dict) – The output of your data iterable, normally a DataLoader.

  • batch_idx (int) – The index of this batch.

  • dataloader_idx – The index of the dataloader that produced this batch. (only if multiple dataloaders used)

Return type:

Tensor | None

Returns:

  • Tensor - The loss tensor

  • dict - A dictionary. Can include any keys, but must include the key 'loss'.

  • None - Skip to the next batch.

# if you have one test dataloader:
def test_step(self, batch, batch_idx): ...


# if you have multiple test dataloaders:
def test_step(self, batch, batch_idx, dataloader_idx=0): ...

Examples:

# CASE 1: A single test dataset
def test_step(self, batch, batch_idx):
    x, y = batch

    # implement your own
    out = self(x)
    loss = self.loss(out, y)

    # log 6 example images
    # or generated text... or whatever
    sample_imgs = x[:6]
    grid = torchvision.utils.make_grid(sample_imgs)
    self.logger.experiment.add_image('example_images', grid, 0)

    # calculate acc
    labels_hat = torch.argmax(out, dim=1)
    test_acc = torch.sum(y == labels_hat).item() / (len(y) * 1.0)

    # log the outputs!
    self.log_dict({'test_loss': loss, 'test_acc': test_acc})

If you pass in multiple test dataloaders, test_step() will have an additional argument. We recommend setting the default value of 0 so that you can quickly switch between single and multiple dataloaders.

# CASE 2: multiple test dataloaders
def test_step(self, batch, batch_idx, dataloader_idx=0):
    # dataloader_idx tells you which dataset this is.
    x, y = batch

    # implement your own
    out = self(x)

    if dataloader_idx == 0:
        loss = self.loss0(out, y)
    else:
        loss = self.loss1(out, y)

    # calculate acc
    labels_hat = torch.argmax(out, dim=1)
    acc = torch.sum(y == labels_hat).item() / (len(y) * 1.0)

    # log the outputs separately for each dataloader
    self.log_dict({f"test_loss_{dataloader_idx}": loss, f"test_acc_{dataloader_idx}": acc})

Note

If you don’t need to test you don’t need to implement this method.

Note

When the test_step() is called, the model has been put in eval mode and PyTorch gradients have been disabled. At the end of the test epoch, the model goes back to training mode and gradients are enabled.

training_step(batch, batch_idx)

Here you compute and return the training loss and some additional metrics for e.g. the progress bar or logger.

Parameters:

  • batch (dict) – The output of your data iterable, normally a DataLoader.

  • batch_idx (int) – The index of this batch.

  • dataloader_idx – The index of the dataloader that produced this batch. (only if multiple dataloaders used)

Return type:

Tensor

Returns:

  • Tensor - The loss tensor

  • dict - A dictionary which can include any keys, but must include the key 'loss' in the case of automatic optimization.

  • None - In automatic optimization, this will skip to the next batch (but is not supported for multi-GPU, TPU, or DeepSpeed). For manual optimization, this has no special meaning, as returning the loss is not required.

In this step you’d normally do the forward pass and calculate the loss for a batch. You can also do fancier things like multiple forward passes or something model specific.

Example:

def training_step(self, batch, batch_idx):
    x, y, z = batch
    out = self.encoder(x)
    loss = self.loss(out, x)
    return loss

To use multiple optimizers, you can switch to ‘manual optimization’ and control their stepping:

def __init__(self):
    super().__init__()
    self.automatic_optimization = False


# Multiple optimizers (e.g.: GANs)
def training_step(self, batch, batch_idx):
    opt1, opt2 = self.optimizers()

    # do training_step with encoder
    ...
    opt1.step()
    # do training_step with decoder
    ...
    opt2.step()

Note

When accumulate_grad_batches > 1, the loss returned here will be automatically normalized by accumulate_grad_batches internally.

validation_step(batch, batch_idx)

Operates on a single batch of data from the validation set. In this step you’d might generate examples or calculate anything of interest like accuracy.

Parameters:

  • batch (dict) – The output of your data iterable, normally a DataLoader.

  • batch_idx (int) – The index of this batch.

  • dataloader_idx – The index of the dataloader that produced this batch. (only if multiple dataloaders used)

Return type:

Tensor | None

Returns:

  • Tensor - The loss tensor

  • dict - A dictionary. Can include any keys, but must include the key 'loss'.

  • None - Skip to the next batch.

# if you have one val dataloader:
def validation_step(self, batch, batch_idx): ...


# if you have multiple val dataloaders:
def validation_step(self, batch, batch_idx, dataloader_idx=0): ...

Examples:

# CASE 1: A single validation dataset
def validation_step(self, batch, batch_idx):
    x, y = batch

    # implement your own
    out = self(x)
    loss = self.loss(out, y)

    # log 6 example images
    # or generated text... or whatever
    sample_imgs = x[:6]
    grid = torchvision.utils.make_grid(sample_imgs)
    self.logger.experiment.add_image('example_images', grid, 0)

    # calculate acc
    labels_hat = torch.argmax(out, dim=1)
    val_acc = torch.sum(y == labels_hat).item() / (len(y) * 1.0)

    # log the outputs!
    self.log_dict({'val_loss': loss, 'val_acc': val_acc})

If you pass in multiple val dataloaders, validation_step() will have an additional argument. We recommend setting the default value of 0 so that you can quickly switch between single and multiple dataloaders.

# CASE 2: multiple validation dataloaders
def validation_step(self, batch, batch_idx, dataloader_idx=0):
    # dataloader_idx tells you which dataset this is.
    x, y = batch

    # implement your own
    out = self(x)

    if dataloader_idx == 0:
        loss = self.loss0(out, y)
    else:
        loss = self.loss1(out, y)

    # calculate acc
    labels_hat = torch.argmax(out, dim=1)
    acc = torch.sum(y == labels_hat).item() / (len(y) * 1.0)

    # log the outputs separately for each dataloader
    self.log_dict({f"val_loss_{dataloader_idx}": loss, f"val_acc_{dataloader_idx}": acc})

Note

If you don’t need to validate you don’t need to implement this method.

Note

When the validation_step() is called, the model has been put in eval mode and PyTorch gradients have been disabled. At the end of validation, the model goes back to training mode and gradients are enabled.

class datamint.lightning.trainers.lightning_modules.UNetPPModule(in_channels, num_classes, loss_fn=None, metrics_factories={}, class_names=None, image_size=None, lr=0.0001, encoder_name='resnet34', encoder_weights='imagenet', transform=None)

Bases: SMPSegmentationModule

Segmentation module using the UNet++ architecture from segmentation_models_pytorch.

Parameters:

to (All arguments are forwarded)

:param SMPSegmentationModule.:

Parameters:

  • in_channels (int)

  • num_classes (int)

  • loss_fn (Module | None)

  • metrics_factories (dict[str, Callable[[], Any]])

  • class_names (list[str] | None)

  • image_size (tuple[int, int] | None)

  • lr (float)

  • encoder_name (str)

  • encoder_weights (str | None)

  • transform (BasicTransform | BaseCompose | None)

forward(x)

Same as torch.nn.Module.forward().

Parameters:

  • *args – Whatever you decide to pass into the forward method.

  • **kwargs – Keyword arguments are also possible.

  • x (Tensor)

Return type:

Tensor

Returns:

Your model’s output

class datamint.lightning.trainers.lightning_modules.YOLOXModule(num_classes, model_size='s', conf_thre=0.25, nms_thre=0.45, lr=0.0001, class_names=None, transform=None, **kwargs)

Bases: DatamintLightningModule

Lightning module wrapping YOLOX for anchor-free object detection.

Parameters:

  • num_classes (int) – Number of object classes (excluding background).

  • model_size (str) – YOLOX size variant — one of 'nano', 'tiny', 's', 'm', 'l', 'x'.

  • conf_thre (float) – Objectness × class-confidence threshold for NMS filtering.

  • nms_thre (float) – IoU threshold for non-maximum suppression.

  • lr (float) – AdamW learning rate.

  • class_names (list[str] | None) – Ordered list of class names (index → name) used to convert model output indices back to semantic labels in predict_image. Must match the DetectionDataset._class_map order.

  • transform (Any) – Eval-time albumentations transform applied in predict_image. Set automatically by the trainer; excluded from saved hyperparameters so checkpoint serialization is not affected.

  • kwargs (Any)

configure_optimizers()

Choose what optimizers and learning-rate schedulers to use in your optimization. Normally you’d need one. But in the case of GANs or similar you might have multiple. Optimization with multiple optimizers only works in the manual optimization mode.

Returns:

Any of these 6 options.

  • Single optimizer.

  • List or Tuple of optimizers.

  • Two lists - The first list has multiple optimizers, and the second has multiple LR schedulers (or multiple lr_scheduler_config).

  • Dictionary, with an "optimizer" key, and (optionally) a "lr_scheduler" key whose value is a single LR scheduler or lr_scheduler_config.

  • None - Fit will run without any optimizer.

The lr_scheduler_config is a dictionary which contains the scheduler and its associated configuration. The default configuration is shown below.

lr_scheduler_config = {
    # REQUIRED: The scheduler instance
    "scheduler": lr_scheduler,
    # The unit of the scheduler's step size, could also be 'step'.
    # 'epoch' updates the scheduler on epoch end whereas 'step'
    # updates it after a optimizer update.
    "interval": "epoch",
    # How many epochs/steps should pass between calls to
    # `scheduler.step()`. 1 corresponds to updating the learning
    # rate after every epoch/step.
    "frequency": 1,
    # Metric to monitor for schedulers like `ReduceLROnPlateau`
    "monitor": "val_loss",
    # If set to `True`, will enforce that the value specified 'monitor'
    # is available when the scheduler is updated, thus stopping
    # training if not found. If set to `False`, it will only produce a warning
    "strict": True,
    # If using the `LearningRateMonitor` callback to monitor the
    # learning rate progress, this keyword can be used to specify
    # a custom logged name
    "name": None,
}

When there are schedulers in which the .step() method is conditioned on a value, such as the torch.optim.lr_scheduler.ReduceLROnPlateau scheduler, Lightning requires that the lr_scheduler_config contains the keyword "monitor" set to the metric name that the scheduler should be conditioned on.

Metrics can be made available to monitor by simply logging it using self.log('metric_to_track', metric_val) in your LightningModule.

Note

Some things to know:

  • Lightning calls .backward() and .step() automatically in case of automatic optimization.

  • If a learning rate scheduler is specified in configure_optimizers() with key "interval" (default “epoch”) in the scheduler configuration, Lightning will call the scheduler’s .step() method automatically in case of automatic optimization.

  • If you use 16-bit precision (precision=16), Lightning will automatically handle the optimizer.

  • If you use torch.optim.LBFGS, Lightning handles the closure function automatically for you.

  • If you use multiple optimizers, you will have to switch to ‘manual optimization’ mode and step them yourself.

  • If you need to control how often the optimizer steps, override the optimizer_step() hook.

forward(x)

Same as torch.nn.Module.forward().

Parameters:

  • *args – Whatever you decide to pass into the forward method.

  • **kwargs – Keyword arguments are also possible.

  • x (Tensor)

Return type:

Tensor

Returns:

Your model’s output

on_test_epoch_end()

Called in the test loop at the very end of the epoch.

Return type:

None

on_validation_epoch_end()

Called in the validation loop at the very end of the epoch.

Return type:

None

predict_image(model_input, **kwargs)

Run detection on a list of resources and return boxes per resource.

Parameters:

  • model_input (Any) – List of BaseResource.

  • kwargs (Any)

Return type:

list

Returns:

list[list[BoxAnnotation]] — one inner list per resource.

task_type = 'object_detection'

Semantic task category for this model class. Subclasses should override at the class body level (e.g. task_type = TaskType.IMAGE_SEGMENTATION).

test_step(batch, batch_idx)

Operates on a single batch of data from the test set. In this step you’d normally generate examples or calculate anything of interest such as accuracy.

Parameters:

  • batch (dict) – The output of your data iterable, normally a DataLoader.

  • batch_idx (int) – The index of this batch.

  • dataloader_idx – The index of the dataloader that produced this batch. (only if multiple dataloaders used)

Return type:

None

Returns:

  • Tensor - The loss tensor

  • dict - A dictionary. Can include any keys, but must include the key 'loss'.

  • None - Skip to the next batch.

# if you have one test dataloader:
def test_step(self, batch, batch_idx): ...


# if you have multiple test dataloaders:
def test_step(self, batch, batch_idx, dataloader_idx=0): ...

Examples:

# CASE 1: A single test dataset
def test_step(self, batch, batch_idx):
    x, y = batch

    # implement your own
    out = self(x)
    loss = self.loss(out, y)

    # log 6 example images
    # or generated text... or whatever
    sample_imgs = x[:6]
    grid = torchvision.utils.make_grid(sample_imgs)
    self.logger.experiment.add_image('example_images', grid, 0)

    # calculate acc
    labels_hat = torch.argmax(out, dim=1)
    test_acc = torch.sum(y == labels_hat).item() / (len(y) * 1.0)

    # log the outputs!
    self.log_dict({'test_loss': loss, 'test_acc': test_acc})

If you pass in multiple test dataloaders, test_step() will have an additional argument. We recommend setting the default value of 0 so that you can quickly switch between single and multiple dataloaders.

# CASE 2: multiple test dataloaders
def test_step(self, batch, batch_idx, dataloader_idx=0):
    # dataloader_idx tells you which dataset this is.
    x, y = batch

    # implement your own
    out = self(x)

    if dataloader_idx == 0:
        loss = self.loss0(out, y)
    else:
        loss = self.loss1(out, y)

    # calculate acc
    labels_hat = torch.argmax(out, dim=1)
    acc = torch.sum(y == labels_hat).item() / (len(y) * 1.0)

    # log the outputs separately for each dataloader
    self.log_dict({f"test_loss_{dataloader_idx}": loss, f"test_acc_{dataloader_idx}": acc})

Note

If you don’t need to test you don’t need to implement this method.

Note

When the test_step() is called, the model has been put in eval mode and PyTorch gradients have been disabled. At the end of the test epoch, the model goes back to training mode and gradients are enabled.

training_step(batch, batch_idx)

Here you compute and return the training loss and some additional metrics for e.g. the progress bar or logger.

Parameters:

  • batch (dict) – The output of your data iterable, normally a DataLoader.

  • batch_idx (int) – The index of this batch.

  • dataloader_idx – The index of the dataloader that produced this batch. (only if multiple dataloaders used)

Return type:

Tensor

Returns:

  • Tensor - The loss tensor

  • dict - A dictionary which can include any keys, but must include the key 'loss' in the case of automatic optimization.

  • None - In automatic optimization, this will skip to the next batch (but is not supported for multi-GPU, TPU, or DeepSpeed). For manual optimization, this has no special meaning, as returning the loss is not required.

In this step you’d normally do the forward pass and calculate the loss for a batch. You can also do fancier things like multiple forward passes or something model specific.

Example:

def training_step(self, batch, batch_idx):
    x, y, z = batch
    out = self.encoder(x)
    loss = self.loss(out, x)
    return loss

To use multiple optimizers, you can switch to ‘manual optimization’ and control their stepping:

def __init__(self):
    super().__init__()
    self.automatic_optimization = False


# Multiple optimizers (e.g.: GANs)
def training_step(self, batch, batch_idx):
    opt1, opt2 = self.optimizers()

    # do training_step with encoder
    ...
    opt1.step()
    # do training_step with decoder
    ...
    opt2.step()

Note

When accumulate_grad_batches > 1, the loss returned here will be automatically normalized by accumulate_grad_batches internally.

validation_step(batch, batch_idx)

Operates on a single batch of data from the validation set. In this step you’d might generate examples or calculate anything of interest like accuracy.

Parameters:

  • batch (dict) – The output of your data iterable, normally a DataLoader.

  • batch_idx (int) – The index of this batch.

  • dataloader_idx – The index of the dataloader that produced this batch. (only if multiple dataloaders used)

Return type:

None

Returns:

  • Tensor - The loss tensor

  • dict - A dictionary. Can include any keys, but must include the key 'loss'.

  • None - Skip to the next batch.

# if you have one val dataloader:
def validation_step(self, batch, batch_idx): ...


# if you have multiple val dataloaders:
def validation_step(self, batch, batch_idx, dataloader_idx=0): ...

Examples:

# CASE 1: A single validation dataset
def validation_step(self, batch, batch_idx):
    x, y = batch

    # implement your own
    out = self(x)
    loss = self.loss(out, y)

    # log 6 example images
    # or generated text... or whatever
    sample_imgs = x[:6]
    grid = torchvision.utils.make_grid(sample_imgs)
    self.logger.experiment.add_image('example_images', grid, 0)

    # calculate acc
    labels_hat = torch.argmax(out, dim=1)
    val_acc = torch.sum(y == labels_hat).item() / (len(y) * 1.0)

    # log the outputs!
    self.log_dict({'val_loss': loss, 'val_acc': val_acc})

If you pass in multiple val dataloaders, validation_step() will have an additional argument. We recommend setting the default value of 0 so that you can quickly switch between single and multiple dataloaders.

# CASE 2: multiple validation dataloaders
def validation_step(self, batch, batch_idx, dataloader_idx=0):
    # dataloader_idx tells you which dataset this is.
    x, y = batch

    # implement your own
    out = self(x)

    if dataloader_idx == 0:
        loss = self.loss0(out, y)
    else:
        loss = self.loss1(out, y)

    # calculate acc
    labels_hat = torch.argmax(out, dim=1)
    acc = torch.sum(y == labels_hat).item() / (len(y) * 1.0)

    # log the outputs separately for each dataloader
    self.log_dict({f"val_loss_{dataloader_idx}": loss, f"val_acc_{dataloader_idx}": acc})

Note

If you don’t need to validate you don’t need to implement this method.

Note

When the validation_step() is called, the model has been put in eval mode and PyTorch gradients have been disabled. At the end of validation, the model goes back to training mode and gradients are enabled.