Merge pull request #89 from orobix/feature/upgrade-timm

Upgrade timm

CHANGELOG.md

@@ -2,6 +2,25 @@
 # Changelog
 All notable changes to this project will be documented in this file.
 
+### [1.4.0]
+
+#### Added
+
+- Add new backbones for classification
+- Add parameter to save a model summary for sklearn based classification tasks
+- Add results csv file for anomaly detection task
+- Add a way to freeze backbone layers by index for the finetuning task
+
+#### Updated
+
+- Update timm requirements to 0.9.12
+
+#### Fixed
+
+- Fix ModelSignatureWrapper not returing the correct instance when cpu, to and half functions are called
+- Fix failure in model logging on mlflow whe half precision is used
+
+
 ### [1.3.8]
 
 #### Updated

docs/tutorials/examples/sklearn_classification.md

@@ -150,6 +150,7 @@ task:
   automatic_batch_size:
     starting_batch_size: 1024
     disable: true
+  save_model_summary: false
   output:
     folder: classification_experiment
     report: true
@@ -161,6 +162,7 @@ This will train a logistic regression classifier using a resnet18 backbone, resi
 The backbone (in torchscript and pytorch format) will be saved along with the classifier. `test_full_data` is used to specify if a final test should be performed on all the data (after training on the training and validation datasets).
 
 Optionally it's possible to enable the automatic batch size finder by setting `automatic_batch_size.disable` to `false`. This will try to find the maximum batch size that can be used on the given device without running out of memory. The `starting_batch_size` parameter is used to specify the starting batch size to use for the search, the algorithm will start from this value and will try to divide it by two until it doesn't run out of memory.
+Finally, the `save_model_summary` parameter can be used to save the backbone information in a text file called `model_summary.txt` located in the root of the output folder.
 
 ### Run
 

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "quadra"
-version = "1.3.8"
+version = "1.4.0"
 description = "Deep Learning experiment orchestration library"
 authors = [
   { name = "Alessandro Polidori", email = "[email protected]" },
@@ -64,10 +64,12 @@ dependencies = [
   "scikit-multilearn==0.2.*",
   "tripy==1.0.*",
   "h5py==3.8.*",
-  "timm==0.6.12", # required by smp
-  "segmentation-models-pytorch==0.3.2",
-  "anomalib@git+https://github.com/orobix/[email protected]+obx.1.2.7",
+  "timm==0.9.12",
+  # Currently the only version of smp supporting timm 0.9.12 is the following
+  "segmentation-models-pytorch@git+https://github.com/qubvel/segmentation_models.pytorch@7b381f899ed472a477a89d381689caf535b5d0a6",
+  "anomalib@git+https://github.com/orobix/[email protected]+obx.1.2.9",
   "xxhash==3.2.*",
+  "torchinfo==1.8.*",
 ]
 
 [project.optional-dependencies]
@@ -121,7 +123,7 @@ repository = "https://github.com/orobix/quadra"
 
 # Adapted from https://realpython.com/pypi-publish-python-package/#version-your-package
 [tool.bumpver]
-current_version = "1.3.8"
+current_version = "1.4.0"
 version_pattern = "MAJOR.MINOR.PATCH"
 commit_message = "build: Bump version {old_version} -> {new_version}"
 commit = true

quadra/__init__.py

@@ -1,4 +1,4 @@
-__version__ = "1.3.8"
+__version__ = "1.4.0"
 
 
 def get_version():

quadra/configs/backbone/caformer_m36.yaml

@@ -0,0 +1,8 @@
+model:
+  _target_: quadra.models.classification.TimmNetworkBuilder
+  model_name: caformer_m36.sail_in22k_ft_in1k
+  pretrained: true
+  freeze: false
+metadata:
+  input_size: 224
+  output_dim: 576

quadra/configs/backbone/caformer_s36.yaml

@@ -0,0 +1,8 @@
+model:
+  _target_: quadra.models.classification.TimmNetworkBuilder
+  model_name: caformer_s36.sail_in22k_ft_in1k
+  pretrained: true
+  freeze: false
+metadata:
+  input_size: 224
+  output_dim: 512

quadra/configs/backbone/convnextv2_base.yaml

@@ -0,0 +1,8 @@
+model:
+  _target_: quadra.models.classification.TimmNetworkBuilder
+  model_name: convnextv2_base.fcmae
+  pretrained: true
+  freeze: false
+metadata:
+  input_size: 224
+  output_dim: 1024

quadra/configs/backbone/convnextv2_femto.yaml

@@ -0,0 +1,8 @@
+model:
+  _target_: quadra.models.classification.TimmNetworkBuilder
+  model_name: convnextv2_femto.fcmae
+  pretrained: true
+  freeze: false
+metadata:
+  input_size: 224
+  output_dim: 384

quadra/configs/backbone/convnextv2_tiny.yaml

@@ -0,0 +1,8 @@
+model:
+  _target_: quadra.models.classification.TimmNetworkBuilder
+  model_name: convnextv2_tiny.fcmae
+  pretrained: true
+  freeze: false
+metadata:
+  input_size: 224
+  output_dim: 768

quadra/configs/backbone/dinov2_vitb14.yaml

@@ -0,0 +1,12 @@
+model:
+  _target_: quadra.models.classification.TorchHubNetworkBuilder
+  repo_or_dir: facebookresearch/dinov2
+  model_name: dinov2_vitb14
+  pretrained: true
+  freeze: false
+  hyperspherical: false
+metadata:
+  input_size: 224
+  output_dim: 768
+  patch_size: 14
+  nb_heads: 12

quadra/configs/backbone/dinov2_vits14.yaml

@@ -0,0 +1,12 @@
+model:
+  _target_: quadra.models.classification.TorchHubNetworkBuilder
+  repo_or_dir: facebookresearch/dinov2
+  model_name: dinov2_vits14
+  pretrained: true
+  freeze: false
+  hyperspherical: false
+metadata:
+  input_size: 224
+  output_dim: 384
+  patch_size: 14
+  nb_heads: 6

quadra/configs/backbone/efficientnet_b0.yaml

@@ -0,0 +1,8 @@
+model:
+  _target_: quadra.models.classification.TimmNetworkBuilder
+  model_name: tf_efficientnet_b0.ns_jft_in1k
+  pretrained: true
+  freeze: false
+metadata:
+  input_size: 224
+  output_dim: 1280

quadra/configs/backbone/efficientnet_b1.yaml

@@ -0,0 +1,8 @@
+model:
+  _target_: quadra.models.classification.TimmNetworkBuilder
+  model_name: tf_efficientnet_b1.ns_jft_in1k
+  pretrained: true
+  freeze: false
+metadata:
+  input_size: 224
+  output_dim: 1280

quadra/configs/backbone/efficientnet_b2.yaml

@@ -0,0 +1,8 @@
+model:
+  _target_: quadra.models.classification.TimmNetworkBuilder
+  model_name: tf_efficientnet_b2.ns_jft_in1k
+  pretrained: true
+  freeze: false
+metadata:
+  input_size: 260
+  output_dim: 1408

quadra/configs/backbone/efficientnet_b3.yaml

@@ -0,0 +1,8 @@
+model:
+  _target_: quadra.models.classification.TimmNetworkBuilder
+  model_name: tf_efficientnet_b3.ns_jft_in1k
+  pretrained: true
+  freeze: false
+metadata:
+  input_size: 300
+  output_dim: 1536

quadra/configs/backbone/resnet18.yaml

@@ -1,6 +1,6 @@
 model:
   _target_: quadra.models.classification.TimmNetworkBuilder
-  model_name: resnet18
+  model_name: resnet18.tv_in1k # Use torchvision weights
   pretrained: true
   freeze: false
 metadata:

quadra/configs/backbone/tiny_vit_21m_224.yaml

@@ -0,0 +1,9 @@
+model:
+  _target_: quadra.models.classification.TimmNetworkBuilder
+  model_name: tiny_vit_21m_224.dist_in22k_ft_in1k
+  pretrained: true
+  freeze: false
+metadata:
+  input_size: 224
+  output_dim: 576
+  num_heads: 18 # Is it correct?

quadra/configs/experiment/base/classification/classification.yaml

@@ -49,7 +49,6 @@ backbone:
   freeze_parameters_name:
 
 trainer:
-  precision: 32
   max_epochs: 200
   check_val_every_n_epoch: 1
   log_every_n_steps: 1

quadra/configs/model/anomalib/padim.yaml

@@ -11,7 +11,7 @@ dataset:
 model:
   name: padim
   input_size: [224, 224]
-  backbone: resnet18
+  backbone: resnet18.tv_in1k
   layers:
     - layer1
     - layer2

quadra/configs/model/anomalib/patchcore.yaml

@@ -10,7 +10,7 @@ dataset:
 
 model:
   name: patchcore
-  backbone: wide_resnet50_2
+  backbone: resnet18.tv_in1k
   layers:
     - layer2
     - layer3

quadra/configs/task/sklearn_classification.yaml

@@ -3,6 +3,7 @@ device: cuda:0
 automatic_batch_size:
   starting_batch_size: 1024
   disable: true
+save_model_summary: false
 output:
   folder: classification_experiment
   report: true

quadra/datamodules/classification.py

@@ -9,7 +9,7 @@
 import torch
 from sklearn.model_selection import train_test_split
 from skmultilearn.model_selection import iterative_train_test_split
-from timm.data.parsers.parser_image_folder import find_images_and_targets
+from timm.data.readers.reader_image_folder import find_images_and_targets
 from torch.utils.data import DataLoader
 
 from quadra.datamodules.base import BaseDataModule

quadra/datasets/anomaly.py

@@ -135,7 +135,9 @@ def make_anomaly_dataset(
     samples_list = [
         (str(path),) + filename.parts[-3:]
         for filename in path.glob("**/*")
-        if filename.is_file() and os.path.splitext(filename)[-1].lower() in IMAGE_EXTENSIONS
+        if filename.is_file()
+        and os.path.splitext(filename)[-1].lower() in IMAGE_EXTENSIONS
+        and ".ipynb_checkpoints" not in str(filename)
     ]
 
     if len(samples_list) == 0:

quadra/models/base.py

@@ -6,10 +6,15 @@
 import torch
 from torch import nn
 
+from quadra.utils.logger import get_logger
+
+log = get_logger(__name__)
+
 
 class ModelSignatureWrapper(nn.Module):
     """Model wrapper used to retrieve input shape. It can be used as a decorator of nn.Module, the first call to the
     forward method will retrieve the input shape and store it in the input_shapes attribute.
+    It will also save the model summary in a file called model_summary.txt in the current working directory.
     """
 
     def __init__(self, model: nn.Module):
@@ -24,16 +29,13 @@ def __init__(self, model: nn.Module):
             self.instance = self.instance.instance
 
     def forward(self, *args: Any, **kwargs: Any) -> torch.Tensor:
-        """Retrieve the input shape and forward the model."""
+        """Retrieve the input shape and forward the model, if the input shape is already retrieved it will just forward
+        the model.
+        """
         if self.input_shapes is None and not self.disable:
             try:
                 self.input_shapes = self._get_input_shapes(*args, **kwargs)
             except Exception:
-                # Avoid circular import
-                # pylint: disable=import-outside-toplevel
-                from quadra.utils.utils import get_logger  # noqa
-
-                log = get_logger(__name__)
                 log.warning(
                     "Failed to retrieve input shapes after forward! To export the model you'll need to "
                     "provide the input shapes manually setting the config.export.input_shapes parameter! "
@@ -46,7 +48,21 @@ def forward(self, *args: Any, **kwargs: Any) -> torch.Tensor:
 
     def to(self, *args, **kwargs):
         """Handle calls to to method returning the underlying model."""
-        return ModelSignatureWrapper(self.instance.to(*args, **kwargs))
+        self.instance = self.instance.to(*args, **kwargs)
+
+        return self
+
+    def half(self, *args, **kwargs):
+        """Handle calls to to method returning the underlying model."""
+        self.instance = self.instance.half(*args, **kwargs)
+
+        return self
+
+    def cpu(self, *args, **kwargs):
+        """Handle calls to to method returning the underlying model."""
+        self.instance = self.instance.cpu(*args, **kwargs)
+
+        return self
 
     def _get_input_shapes(self, *args: Any, **kwargs: Any) -> list[Any]:
         """Retrieve the input shapes from the input. Inputs will be in the same order as the forward method
@@ -115,7 +131,17 @@ def __setattr__(self, name: str, value: torch.Tensor | nn.Module) -> None:
             setattr(self.instance, name, value)
 
     def __getattribute__(self, __name: str) -> Any:
-        if __name in ["instance", "input_shapes", "__dict__", "forward", "_get_input_shapes", "_get_input_shape", "to"]:
+        if __name in [
+            "instance",
+            "input_shapes",
+            "__dict__",
+            "forward",
+            "_get_input_shapes",
+            "_get_input_shape",
+            "to",
+            "half",
+            "cpu",
+        ]:
             return super().__getattribute__(__name)
 
         return getattr(self.instance, __name)