From c559d8efb26d60f687870787b868941f7282315b Mon Sep 17 00:00:00 2001
From: Manon blanco <blanco@teklia.com>
Date: Tue, 13 Jun 2023 06:20:36 +0000
Subject: [PATCH] Remove normalize parameter from training configuration
---
dan/manager/ocr.py | 5 ++---
dan/ocr/document/train.py | 1 -
docs/usage/train/parameters.md | 1 -
tests/conftest.py | 1 -
4 files changed, 2 insertions(+), 6 deletions(-)
diff --git a/dan/manager/ocr.py b/dan/manager/ocr.py
index d1562780..b70b9364 100644
--- a/dan/manager/ocr.py
+++ b/dan/manager/ocr.py
@@ -112,9 +112,8 @@ class OCRDataset(GenericDataset):
)
sample["img"] = cv2.resize(sample["img"], (new_w, new_h))
- # Normalization if requested
- if "normalize" in self.params["config"] and self.params["config"]["normalize"]:
- sample["img"] = (sample["img"] - self.mean) / self.std
+ # Normalization
+ sample["img"] = (sample["img"] - self.mean) / self.std
sample["img_reduced_shape"] = np.ceil(
sample["img"].shape / self.reduce_dims_factor
diff --git a/dan/ocr/document/train.py b/dan/ocr/document/train.py
index 4745fb97..a3bbba55 100644
--- a/dan/ocr/document/train.py
+++ b/dan/ocr/document/train.py
@@ -113,7 +113,6 @@ def get_config():
"add_eot",
"add_sot",
], # add end-of-transcription and start-of-transcription tokens in labels
- "normalize": True, # Normalize with mean and variance of training dataset
"preprocessings": [
{
"type": "to_RGB",
diff --git a/docs/usage/train/parameters.md b/docs/usage/train/parameters.md
index 7447e082..9c82931b 100644
--- a/docs/usage/train/parameters.md
+++ b/docs/usage/train/parameters.md
@@ -19,7 +19,6 @@ All hyperparameters are specified and editable in the training scripts (meaning
| `dataset_params.config.padding_value` | Image padding value. | `int` | `0` |
| `dataset_params.config.padding_token` | Transcription padding value. | `int` | `None` |
| `dataset_params.config.constraints` | Whether to add end-of-transcription and start-of-transcription tokens in labels. | `list` | `["add_eot", "add_sot"]` |
-| `dataset_params.config.normalize` | Normalize with mean and variance of training dataset. | `bool` | `True` |
| `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)) |
diff --git a/tests/conftest.py b/tests/conftest.py
index 7febae99..854de3c0 100644
--- a/tests/conftest.py
+++ b/tests/conftest.py
@@ -76,7 +76,6 @@ def training_config():
"add_eot",
"add_sot",
], # add end-of-transcription and start-of-transcription tokens in labels
- "normalize": True, # Normalize with mean and variance of training dataset
"preprocessings": [
{
"type": "to_RGB",
--
GitLab