Skip to content
Snippets Groups Projects

Compare revisions

Changes are shown as if the source revision was being merged into the target revision. Learn more about comparing revisions.

Source

Select target project
No results found

Target

Select target project
  • atr/dan
1 result
Show changes
Hide whitespace changes
Inline Side-by-side
Showing
with 262 additions and 145 deletions
docs/assets/augmentations/document_shearx.png 0 → 100644
View file @ 5a7dd7ef
docs/assets/augmentations/document_shearx.png

374 KiB

docs/assets/augmentations/line_color_jitter.png 0 → 100644
View file @ 5a7dd7ef
docs/assets/augmentations/line_color_jitter.png

85.9 KiB

docs/assets/augmentations/line_downscale.png 0 → 100644
View file @ 5a7dd7ef
docs/assets/augmentations/line_downscale.png

86.9 KiB

docs/assets/augmentations/line_dropout.png 0 → 100644
View file @ 5a7dd7ef
docs/assets/augmentations/line_dropout.png

83.5 KiB

docs/assets/augmentations/line_elastic.png 0 → 100644
View file @ 5a7dd7ef
docs/assets/augmentations/line_elastic.png

113 KiB

docs/assets/augmentations/line_erosion_dilation.png 0 → 100644
View file @ 5a7dd7ef
docs/assets/augmentations/line_erosion_dilation.png

90.2 KiB

docs/assets/augmentations/line_full_pipeline.png 0 → 100644
View file @ 5a7dd7ef
docs/assets/augmentations/line_full_pipeline.png

329 KiB

docs/assets/augmentations/line_gaussian_blur.png 0 → 100644
View file @ 5a7dd7ef
docs/assets/augmentations/line_gaussian_blur.png

86.1 KiB

docs/assets/augmentations/line_gaussian_noise.png 0 → 100644
View file @ 5a7dd7ef
docs/assets/augmentations/line_gaussian_noise.png

105 KiB

docs/assets/augmentations/line_grayscale.png 0 → 100644
View file @ 5a7dd7ef
docs/assets/augmentations/line_grayscale.png

30 KiB

docs/assets/augmentations/line_perspective.png 0 → 100644
View file @ 5a7dd7ef
docs/assets/augmentations/line_perspective.png

82 KiB

docs/assets/augmentations/line_piecewise.png 0 → 100644
View file @ 5a7dd7ef
docs/assets/augmentations/line_piecewise.png

86.9 KiB

docs/assets/augmentations/line_sharpen.png 0 → 100644
View file @ 5a7dd7ef
docs/assets/augmentations/line_sharpen.png

90 KiB

docs/assets/augmentations/line_shearx.png 0 → 100644
View file @ 5a7dd7ef
docs/assets/augmentations/line_shearx.png

84.6 KiB

docs/usage/train/augmentation.md 0 → 100644
View file @ 5a7dd7ef
# Data augmentation transforms
This page lists data augmentation transforms used in DAN.
## Individual augmentation transforms
### Elastic Transform
| | Elastic Transform |
| ---------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| Description | This transformation applies local distortions that rotate characters locally. |
| Comments | The impact of this transformation is mostly visible on documents, not so much on lines. Results are comparable to the original DAN implementation. |
| Documentation | See the [`albumentations` documentation](https://albumentations.ai/docs/api_reference/augmentations/geometric/transforms/#albumentations.augmentations.geometric.transforms.ElasticTransform). |
| Examples | ![](../../assets/augmentations/line_elastic.png) ![](../../assets/augmentations/document_elastic.png) |
| CPU time (seconds/10 images) | 0.44 (3013x128 pixels) / 0.86 (1116x581 pixels) |
### PieceWise Affine
| | PieceWise Affine |
| ---------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| Description | This transformation also applies local distortions but with a larger grid than ElasticTransform. |
| Comments | This transformation is very slow. It is a new transform that was not in the original DAN implementation. |
| Documentation | See the [`albumentations` documentation](https://albumentations.ai/docs/api_reference/augmentations/geometric/transforms/#albumentations.augmentations.geometric.transforms.PiecewiseAffine). |
| Examples | ![](../../assets/augmentations/line_piecewise.png) ![](../../assets/augmentations/document_piecewise.png) |
| CPU time (seconds/10 images) | 2.92 (3013x128 pixels) / 3.76 (1116x581 pixels) |
### Dilation Erosion
| | Dilation & Erosion |
| ---------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
| Description | This transformation makes the pen stroke thicker or thinner. |
| Comments | The `RandomDilationErosion` class randomly selects a kernel size and applies a dilation or an erosion to the image. It relies on opencv and is similar to the original DAN implementation. |
| Documentation | See the [`opencv` documentation](https://docs.opencv.org/3.4/db/df6/tutorial_erosion_dilatation.html). |
| Examples | ![](../../assets/augmentations/line_erosion_dilation.png) ![](../../assets/augmentations/document_erosion_dilation.png) |
| CPU time (seconds/10 images) | 0.02 (3013x128 pixels) / 0.03 (1116x581 pixels) |
### Sharpen
| | Sharpen |
| ---------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| Description | This transformation makes the image sharper. |
| Comments | Similar to the original DAN implementation. |
| Documentation | See the [`albumentations` documentation](https://albumentations.ai/docs/api_reference/augmentations/transforms/#albumentations.augmentations.transforms.Sharpen). |
| Examples | ![](../../assets/augmentations/line_sharpen.png) ![](../../assets/augmentations/document_sharpen.png) |
| CPU time (seconds/10 images) | 0.02 (3013x128 pixels) / 0.04 (1116x581 pixels) |
### Color Jittering
| | Color Jittering |
| ---------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| Description | This transformation alters the colors of the image. |
| Comments | Similar to the original DAN implementation. |
| Documentation | See the [`albumentations` documentation](https://albumentations.ai/docs/api_reference/augmentations/transforms/#albumentations.augmentations.transforms.ColorJitter). |
| Examples | ![](../../assets/augmentations/line_color_jitter.png) ![](../../assets/augmentations/document_color_jitter.png) |
| CPU time (seconds/10 images) | 0.03 (3013x128 pixels) / 0.04 (1116x581 pixels) |
### Gaussian Noise
| | Gaussian Noise |
| ---------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| Description | This transformation adds Gaussian noise to the image. |
| Comments | The noise from the original DAN implementation is more uniform. |
| Documentation | See the [`albumentations` documentation](https://albumentations.ai/docs/api_reference/augmentations/transforms/#albumentations.augmentations.transforms.GaussianNoise). |
| Examples | ![](../../assets/augmentations/line_gaussian_noise.png) ![](../../assets/augmentations/document_gaussian_noise.png) |
| CPU time (seconds/10 images) | 0.29 (3013x128 pixels) / 0.53 (1116x581 pixels) |
### Gaussian Blur
| | Gaussian Blur |
| ---------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| Description | This transformation blurs the image. |
| Comments | Similar to the original DAN implementation. |
| Documentation | See the [`albumentations` documentation](https://albumentations.ai/docs/api_reference/augmentations/transforms/#albumentations.augmentations.transforms.GaussianBlur). |
| Examples | ![](../../assets/augmentations/line_gaussian_blur.png) ![](../../assets/augmentations/document_gaussian_blur.png) |
| CPU time (seconds/10 images) | 0.01 (3013x128 pixels) / 0.02 (1116x581 pixels) |
### Random Perspective
| | Random Perspective |
| ---------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| Description | This transformation changes the perspective from which the photo is taken. |
| Comments | Similar to the original DAN implementation. |
| Documentation | See the [`albumentations` documentation](https://albumentations.ai/docs/api_reference/augmentations/transforms/#albumentations.augmentations.transforms.Perspective). |
| Examples | ![](../../assets/augmentations/line_perspective.png) ![](../../assets/augmentations/document_perspective.png) |
| CPU time (seconds/10 images) | 0.05 (3013x128 pixels) / 0.05 (1116x581 pixels) |
### Shearing (x-axis)
| | Shearing (x-axis) |
| ---------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
| Description | This transformation changes the slant of the text on the image. |
| Comments | New transform that was not in the original DAN implementation. |
| Documentation | See the [`albumentations` documentation](https://albumentations.ai/docs/api_reference/augmentations/geometric/transforms/#albumentations.augmentations.geometric.transforms.Affine). |
| Examples | ![](../../assets/augmentations/line_shearx.png) ![](../../assets/augmentations/document_shearx.png) |
| CPU time (seconds/10 images) | 0.05 (3013x128 pixels) / 0.04 (1116x581 pixels) |
### Coarse Dropout
| | Coarse Dropout |
| ---------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| Description | This transformation adds dropout on the image, turning small patches into black pixels. |
| Comments | It is a new transform that was not in the original DAN implementation. |
| Documentation | See the [`albumentations` documentation](https://albumentations.ai/docs/api_reference/augmentations/dropout/coarse_dropout/#coarsedropout-augmentation-augmentationsdropoutcoarse_dropout). |
| Examples | ![](../../assets/augmentations/line_dropout.png) ![](../../assets/augmentations/document_dropout.png) |
| CPU time (seconds/10 images) | 0.02 (3013x128 pixels) / 0.02 (1116x581 pixels) |
### Downscale
| | Downscale |
| ---------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| Description | This transformation downscales the image by a random factor. |
| Comments | It is a new transform that was not in the original DAN implementation. |
| Documentation | See the [`albumentations` documentation](https://albumentations.ai/docs/api_reference/augmentations/transforms/#albumentations.augmentations.transforms.Downscale). |
| Examples | ![](../../assets/augmentations/line_downscale.png) ![](../../assets/augmentations/document_downscale.png) |
| CPU time (seconds/10 images) | 0.03 (3013x128 pixels) / 0.03 (1116x581 pixels) |
### Grayscale
| | Grayscale |
| ---------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| Description | This transformation transforms an RGB image into grayscale. |
| Comments | It is a new transform that was not in the original DAN implementation. |
| Documentation | See the [`albumentations` documentation](https://albumentations.ai/docs/api_reference/augmentations/transforms/#albumentations.augmentations.transforms.ToGray). |
| Examples | ![](../../assets/augmentations/line_grayscale.png) ![](../../assets/augmentations/document_grayscale.png) |
| CPU time (seconds/10 images) | 0.02 (3013x128 pixels) / 0.02 (1116x581 pixels) |
## Full augmentation pipeline
* Data augmentation is applied with a probability of 0.9.
* In this case, two transformations are randomly selected to be applied.
* `ElasticTransform` and `PieceWiseAffine` cannot be applied on the same image.
* Reproducibility is possible by setting `random.seed` and `np.random.seed` (already done in `dan/ocr/document/train.py`)
* Examples with new pipeline:
![](../../assets/augmentations/line_full_pipeline.png)
![](../../assets/augmentations/document_full_pipeline.png)
![](../../assets/augmentations/document_full_pipeline_2.png)
docs/usage/train/index.md
View file @ 5a7dd7ef
......@@ -21,3 +21,4 @@ To train DAN on lines, run `teklia-dan train document` with a line dataset.
## Additional page
* [Jean Zay tutorial](jeanzay.md)
* [Data augmentation](augmentation.md)
docs/usage/train/parameters.md
View file @ 5a7dd7ef
......@@ -3,65 +3,82 @@ All hyperparameters are specified and editable in the training scripts (meaning
## Dataset parameters
| Parameter | Description | Type | Default |
| --------------------------------------- | -------------------------------------------------------------------------------------- | ------------ | ---------------------------------------------- |
| `dataset_name` | Name of the dataset. | `str` | |
| `dataset_level` | Level of the dataset. Should be named after the element type. | `str` | |
| `dataset_variant` | Variant of the dataset. Usually empty for HTR datasets, `"_sem"` for HTR+NER datasets. | `str` | |
| `dataset_path` | Path to the dataset. | `str` | |
| `dataset_params.config.dataset_manager` | Dataset manager class. | custom class | `OCRDatasetManager` |
| `dataset_params.config.dataset_class` | Dataset class. | custom class | `OCRDataset` |
| `dataset_params.config.datasets` | Dataset dictionary with the dataset name as key and dataset path as value. | `dict` | |
| `dataset_params.config.load_in_memory` | Load all images in CPU memory. | `str` | `True` |
| `dataset_params.config.worker_per_gpu` | Number of parallel processes per gpu for data loading. | `int` | `4` |
| `dataset_params.config.preprocessings` | List of pre-processing functions to apply to input images. | `list` | (see [dedicated section](#data-preprocessing)) |
| `dataset_params.config.augmentation` | Configuration for data augmentation. | `dict` | (see [dedicated section](#data-augmentation)) |
| Parameter | Description | Type | Default |
| --------------------------------------- | -------------------------------------------------------------------------------------- | ------------ | ---------------------------------------------------- |
| `dataset_name` | Name of the dataset. | `str` | |
| `dataset_level` | Level of the dataset. Should be named after the element type. | `str` | |
| `dataset_variant` | Variant of the dataset. Usually empty for HTR datasets, `"_sem"` for HTR+NER datasets. | `str` | |
| `dataset_path` | Path to the dataset. | `str` | |
| `dataset_params.config.dataset_manager` | Dataset manager class. | custom class | `OCRDatasetManager` |
| `dataset_params.config.dataset_class` | Dataset class. | custom class | `OCRDataset` |
| `dataset_params.config.datasets` | Dataset dictionary with the dataset name as key and dataset path as value. | `dict` | |
| `dataset_params.config.load_in_memory` | Load all images in CPU memory. | `str` | `True` |
| `dataset_params.config.worker_per_gpu` | Number of parallel processes per gpu for data loading. | `int` | `4` |
| `dataset_params.config.preprocessings` | List of pre-processing functions to apply to input images. | `list` | (see [dedicated section](#data-preprocessing)) |
| `dataset_params.config.augmentation` | Whether to use data augmentation on the training set. | `bool` | `True` (see [dedicated section](#data-augmentation)) |
### Data preprocessing
Preprocessing is applied before training the network (see `dan/manager/dataset.py`).
The following transformations are implemented:
Preprocessing is applied before training the network (see `dan/manager/dataset.py`). The list of accepted transforms is defined in `dan/transforms.py`:
* Convert to grayscale
```py
{
"type": "to_grayscaled"
}
class Preprocessing(Enum):
# If the image is bigger than the given size, resize it while keeping the original ratio
MaxResize = "max_resize"
# Resize the height to a fixed value while keeping the original ratio
FixedHeightResize = "fixed_height_resize"
# Resize the width to a fixed value while keeping the original ratio
FixedWidthResize = "fixed_width_resize"
```
* Convert to RGB
Usage:
* Resize to a fixed height
```py
{
"type": "to_RGB"
}
[
{
"type": Preprocessing.FixedHeightResize,
"fixed_height": 1500,
}
]
```
* Resize to a fixed height
* Resize to a fixed width
```py
{
"type": "fixed_height",
"fixed_height": 1000,
}
[
{
"type": Preprocessing.FixedWidthResize,
"fixed_height": 1500,
}
]
```
* Resize to a maximum size
* Resize to a maximum size (only if the image is bigger than the given size)
```py
{
"type": "resize",
"keep_ratio": True,
"max_height": 1000,
"max_width": None,
}
[
{
"type": Preprocessing.MaxResize,
"max_height": 2000,
"max_width": 2000,
}
]
```
Multiple transformations can be combined. For example, to resize an image to a fixed height of 1000 pixels and convert images to RGB, use the following configuration in `dataset_params.config.preprocessings`:
* Combine these pre-processings
```py
[
{
"type": "fixed_height",
"fixed_height": 1000
"type": Preprocessing.FixedHeightResize,
"fixed_height": 2000,
},
{
"type": "to_RGB"
"type": Preprocessing.FixedWidthResize,
"fixed_width": 2000,
}
]
```
......@@ -70,83 +87,41 @@ Multiple transformations can be combined. For example, to resize an image to a f
Augmentation transformations are applied on-the-fly during training to artificially increase data variability.
The following transformations are implemented in `dan/transforms.py`:
* Color inversion
* Dilation and erosion
* Elastic distortion
* Reducing interline spacing
* Gaussian blur
* Gaussian noise
DAN also takes advantage of [transforms from torchvision](https://pytorch.org/vision/stable/transforms.html):
* ColorJitter
* GaussianBlur
* RandomCrop
* RandomPerspective
The following configuration is used by default when using the `teklia-dan train document` command. Data augmentation is applied with a probability of 0.9, and each transformation has a 0.1 probability to be used.
DAN takes advantage of transforms from [albumentations](https://albumentations.ai/).
The following configuration is used by default when using the `teklia-dan train document` command. Data augmentation is applied with a probability of 0.9. In this case, two transformations are randomly selected to be applied.
```py
{
"order": "random",
"proba": 0.9,
"augmentations": [
{
"type": "perspective",
"proba": 0.1,
"min_factor": 0,
"max_factor": 0.4,
},
{
"type": "elastic_distortion",
"proba": 0.1,
"min_alpha": 0.5,
"max_alpha": 1,
"min_sigma": 1,
"max_sigma": 10,
"min_kernel_size": 3,
"max_kernel_size": 9,
},
{
"type": "dilation_erosion",
"proba": 0.1,
"min_kernel": 1,
"max_kernel": 3,
"iterations": 1,
},
{
"type": "color_jittering",
"proba": 0.1,
"factor_hue": 0.2,
"factor_brightness": 0.4,
"factor_contrast": 0.4,
"factor_saturation": 0.4,
},
{
"type": "gaussian_blur",
"proba": 0.1,
"min_kernel": 3,
"max_kernel": 5,
"min_sigma": 3,
"max_sigma": 5,
},
{
"type": "gaussian_noise",
"proba": 0.1,
"std": 0.5,
},
{
"type": "sharpen",
"proba": 0.1,
"min_alpha": 0,
"max_alpha": 1,
"min_strength": 0,
"max_strength": 1,
},
],
}
transforms = SomeOf(
[
Perspective(scale=(0.05, 0.09), fit_output=True),
GaussianBlur(sigma_limit=2.5),
GaussNoise(var_limit=50**2),
ColorJitter(contrast=0.2, brightness=0.2, saturation=0.2, hue=0.2),
OneOf(
[
ElasticTransform(
alpha=20.0,
sigma=5.0,
alpha_affine=1.0,
border_mode=0,
),
PiecewiseAffine(scale=(0.01, 0.04), nb_rows=1, nb_cols=4),
]
),
Sharpen(alpha=(0.0, 1.0)),
ErosionDilation(min_kernel=1, max_kernel=4, iterations=1),
Affine(shear={"x": (-20, 20), "y": (0, 0)}),
CoarseDropout(),
Downscale(scale_min=0.5, scale_max=0.9, interpolation=INTER_NEAREST),
ToGray(),
],
n=2,
p=0.9,
)
```
For a detailed description of all augmentation transforms, see the [dedicated page](augmentation.md).
## Model parameters
| Name | Description | Type | Default |
......@@ -180,32 +155,32 @@ The following configuration is used by default when using the `teklia-dan train
## Training parameters
| Name | Description | Type | Default |
| ----------------------------------------------------------- | --------------------------------------------------------------------------- | ------------ | ------------------------------------------- |
| `training_params.output_folder` | Directory for checkpoint and results. | `str` | |
| `training_params.max_nb_epochs` | Maximum number of epochs before stopping training. | `int` | `800` |
| `training_params.max_training_time` | Maximum time (in seconds) before stopping training. | `int` | `350000` |
| `training_params.load_epoch` | Model to load. Should be either `"best"` (evaluation) or `last` (training). | `str` | `"last"` |
| `training_params.interval_save_weights` | Step to save weights. Set to `None` to keep only best and last epochs. | `int` | `None` |
| `training_params.batch_size` | Mini-batch size for the training loop. | `int` | `2` |
| `training_params.valid_batch_size` | Mini-batch size for the valdiation loop. | `int` | `4` |
| `training_params.use_ddp` | Whether to use DistributedDataParallel. | `bool` | `False` |
| `training_params.ddp_port` | DDP port. | `int` | `20027` |
| `training_params.use_amp` | Whether to enable automatic mix-precision. | `int` | `torch.cuda.device_count()` |
| `training_params.nb_gpu` | Number of GPUs to train DAN. | `str` | |
| `training_params.optimizers.all.class` | Optimizer class. | custom class | `Adam` |
| `training_params.optimizers.all.args.lr` | Learning rate for the optimizer. | `float` | `0.0001` |
| `training_params.optimizers.all.args.amsgrad` | Whether to use AMSGrad optimization. | custom class | `False` |
| `training_params.lr_schedulers` | Learning rate schedulers. | custom class | `None` |
| `training_params.eval_on_valid` | Whether to evaluate and log metrics on the validation set during training. | `bool` | `True` |
| `training_params.eval_on_valid_interval` | Interval (in epochs) to evaluate during training. | `int` | `5` |
| `training_params.focus_metric` | Metrics to focus on to determine best epoch. | `str` | `cer` |
| `training_params.expected_metric_value` | Best value for the focus metric. Should be either `"high"` or `"low"`. | `low` | `cer` |
| `training_params.set_name_focus_metric` | Dataset to focus on to select best weights. | `str` | |
| `training_params.train_metrics` | List of metrics to compute during training. | `list` | `["loss_ce", "cer", "wer", "wer_no_punct"]` |
| `training_params.train_metrics` | List of metrics to compute during validation. | `list` | `["cer", "wer", "wer_no_punct"]` |
| `training_params.force_cpu` | Whether to train on CPU (for debugging). | `bool` | `False` |
| `training_params.max_char_prediction` | Maximum number of characters to predict. | `int` | `1000` |
| Name | Description | Type | Default |
| ------------------------------------------------------- | --------------------------------------------------------------------------- | ------------ | ------------------------------------------- |
| `training_params.output_folder` | Directory for checkpoint and results. | `str` | |
| `training_params.max_nb_epochs` | Maximum number of epochs before stopping training. | `int` | `800` |
| `training_params.max_training_time` | Maximum time (in seconds) before stopping training. | `int` | `350000` |
| `training_params.load_epoch` | Model to load. Should be either `"best"` (evaluation) or `last` (training). | `str` | `"last"` |
| `training_params.interval_save_weights` | Step to save weights. Set to `None` to keep only best and last epochs. | `int` | `None` |
| `training_params.batch_size` | Mini-batch size for the training loop. | `int` | `2` |
| `training_params.valid_batch_size` | Mini-batch size for the valdiation loop. | `int` | `4` |
| `training_params.use_ddp` | Whether to use DistributedDataParallel. | `bool` | `False` |
| `training_params.ddp_port` | DDP port. | `int` | `20027` |
| `training_params.use_amp` | Whether to enable automatic mix-precision. | `int` | `torch.cuda.device_count()` |
| `training_params.nb_gpu` | Number of GPUs to train DAN. | `str` | |
| `training_params.optimizers.all.class` | Optimizer class. | custom class | `Adam` |
| `training_params.optimizers.all.args.lr` | Learning rate for the optimizer. | `float` | `0.0001` |
| `training_params.optimizers.all.args.amsgrad` | Whether to use AMSGrad optimization. | custom class | `False` |
| `training_params.lr_schedulers` | Learning rate schedulers. | custom class | `None` |
| `training_params.eval_on_valid` | Whether to evaluate and log metrics on the validation set during training. | `bool` | `True` |
| `training_params.eval_on_valid_interval` | Interval (in epochs) to evaluate during training. | `int` | `5` |
| `training_params.focus_metric` | Metrics to focus on to determine best epoch. | `str` | `cer` |
| `training_params.expected_metric_value` | Best value for the focus metric. Should be either `"high"` or `"low"`. | `low` | `cer` |
| `training_params.set_name_focus_metric` | Dataset to focus on to select best weights. | `str` | |
| `training_params.train_metrics` | List of metrics to compute during training. | `list` | `["loss_ce", "cer", "wer", "wer_no_punct"]` |
| `training_params.train_metrics` | List of metrics to compute during validation. | `list` | `["cer", "wer", "wer_no_punct"]` |
| `training_params.force_cpu` | Whether to train on CPU (for debugging). | `bool` | `False` |
| `training_params.max_char_prediction` | Maximum number of characters to predict. | `int` | `1000` |
| `training_params.label_noise_scheduler.min_error_rate` | Minimum ratio of teacher forcing. | `float` | `0.2` |
| `training_params.label_noise_scheduler.max_error_rate` | Maximum ratio of teacher forcing. | `float` | `0.2` |
| `training_params.label_noise_scheduler.total_num_steps` | Number of steps before stopping teacher forcing. | `float` | `5e4` |
......@@ -214,11 +189,12 @@ The following configuration is used by default when using the `teklia-dan train
## MLFlow logging
To log your experiment on MLFlow, you need to:
- install the extra requirements via
```shell
$ pip install .[mlflow]
```
```shell
$ pip install .[mlflow]
```
- update the following arguments:
......
mkdocs.yml
View file @ 5a7dd7ef
......@@ -62,6 +62,7 @@ nav:
- Training:
- usage/train/index.md
- Parameters: usage/train/parameters.md
- Data augmentation: usage/train/augmentation.md
- Jean Zay tutorial: usage/train/jeanzay.md
- Predict: usage/predict.md
- Documentation development: dev/build_docs.md
......
requirements.txt
View file @ 5a7dd7ef
albumentations==1.3.1
arkindex-export==0.1.3
boto3==1.26.124
editdistance==0.6.2
......
tests/conftest.py
View file @ 5a7dd7ef
......@@ -9,7 +9,7 @@ from torch.optim import Adam
from dan.decoder import GlobalHTADecoder
from dan.encoder import FCN_Encoder
from dan.schedulers import exponential_dropout_scheduler
from dan.transforms import aug_config
from dan.transforms import Preprocessing
FIXTURES = Path(__file__).resolve().parent / "data"
......@@ -70,11 +70,12 @@ def training_config():
"load_in_memory": True, # Load all images in CPU memory
"preprocessings": [
{
"type": "to_RGB",
# if grayscaled image, produce RGB one (3 channels with same value) otherwise do nothing
"type": Preprocessing.MaxResize,
"max_width": 2000,
"max_height": 2000,
},
],
"augmentation": aug_config(0.9, 0.1),
"augmentation": True,
},
},
"model_params": {
......