From 7a4af1eb3033511a43533a4b5239e7be6501fb0a Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Sol=C3=A8ne=20Tarride?= <starride@teklia.com>
Date: Mon, 5 Dec 2022 09:51:41 +0000
Subject: [PATCH] monitor probability of synthetic documents
---
dan/manager/metrics.py | 9 ++++++-
dan/manager/ocr.py | 50 +++++++++++++++++++++++----------------
dan/manager/training.py | 3 +++
dan/ocr/document/train.py | 1 +
4 files changed, 42 insertions(+), 21 deletions(-)
diff --git a/dan/manager/metrics.py b/dan/manager/metrics.py
index c9e3a642..36d3a29f 100644
--- a/dan/manager/metrics.py
+++ b/dan/manager/metrics.py
@@ -127,7 +127,13 @@ class MetricManager:
)
if output:
display_values["nb_words"] = np.sum(self.epoch_metrics["nb_words"])
- elif metric_name in ["loss", "loss_ctc", "loss_ce", "syn_max_lines"]:
+ elif metric_name in [
+ "loss",
+ "loss_ctc",
+ "loss_ce",
+ "syn_max_lines",
+ "syn_prob_lines",
+ ]:
value = np.average(
self.epoch_metrics[metric_name],
weights=np.array(self.epoch_metrics["nb_samples"]),
@@ -182,6 +188,7 @@ class MetricManager:
"loss_ce",
"loss",
"syn_max_lines",
+ "syn_prob_lines",
]:
metrics[metric_name] = [
values[metric_name],
diff --git a/dan/manager/ocr.py b/dan/manager/ocr.py
index 43d46894..d097eae3 100644
--- a/dan/manager/ocr.py
+++ b/dan/manager/ocr.py
@@ -284,9 +284,37 @@ class OCRDataset(GenericDataset):
def generate_synthetic_data(self, sample):
config = self.params["config"]["synthetic_data"]
+ proba = self.get_syn_proba_lines()
- if not (config["init_proba"] == config["end_proba"] == 1):
+ if rand() > proba:
+ return sample
+
+ if "mode" in config and config["mode"] == "line_hw_to_printed":
+ sample["img"] = self.generate_typed_text_line_image(sample["label"])
+ return sample
+
+ return self.generate_synthetic_page_sample()
+
+ def get_syn_max_lines(self):
+ config = self.params["config"]["synthetic_data"]
+ if config["curriculum"]:
nb_samples = self.training_info["step"] * self.params["batch_size"]
+ max_nb_lines = min(
+ config["max_nb_lines"],
+ (nb_samples - config["curr_start"]) // config["curr_step"] + 1,
+ )
+ return max(config["min_nb_lines"], max_nb_lines)
+ return config["max_nb_lines"]
+
+ def get_syn_proba_lines(self):
+ config = self.params["config"]["synthetic_data"]
+
+ if config["init_proba"] == config["end_proba"]:
+ return config["init_proba"]
+
+ else:
+ nb_samples = self.training_info["step"] * self.params["batch_size"]
+
if config["start_scheduler_at_max_line"]:
max_step = config["num_steps_proba"]
current_step = max(
@@ -315,25 +343,7 @@ class OCRDataset(GenericDataset):
min(nb_samples, config["num_steps_proba"]),
config["num_steps_proba"],
)
- if rand() > proba:
- return sample
-
- if "mode" in config and config["mode"] == "line_hw_to_printed":
- sample["img"] = self.generate_typed_text_line_image(sample["label"])
- return sample
-
- return self.generate_synthetic_page_sample()
-
- def get_syn_max_lines(self):
- config = self.params["config"]["synthetic_data"]
- if config["curriculum"]:
- nb_samples = self.training_info["step"] * self.params["batch_size"]
- max_nb_lines = min(
- config["max_nb_lines"],
- (nb_samples - config["curr_start"]) // config["curr_step"] + 1,
- )
- return max(config["min_nb_lines"], max_nb_lines)
- return config["max_nb_lines"]
+ return proba
def generate_synthetic_page_sample(self):
config = self.params["config"]["synthetic_data"]
diff --git a/dan/manager/training.py b/dan/manager/training.py
index 682e3091..41ba05cb 100644
--- a/dan/manager/training.py
+++ b/dan/manager/training.py
@@ -1160,6 +1160,9 @@ class Manager(OCRManager):
"syn_max_lines": self.dataset.train_dataset.get_syn_max_lines()
if self.params["dataset_params"]["config"]["synthetic_data"]
else 0,
+ "syn_prob_lines": self.dataset.train_dataset.get_syn_proba_lines()
+ if self.params["dataset_params"]["config"]["synthetic_data"]
+ else 0,
}
return values
diff --git a/dan/ocr/document/train.py b/dan/ocr/document/train.py
index a4d3f5e8..b173ba99 100644
--- a/dan/ocr/document/train.py
+++ b/dan/ocr/document/train.py
@@ -209,6 +209,7 @@ def run():
"cer",
"wer",
"syn_max_lines",
+ "syn_prob_lines",
], # Metrics name for training
"eval_metrics": [
"cer",
--
GitLab