Use a single padding method

Closes #104 (closed)

dan/manager/ocr.py

@@ -6,10 +6,9 @@ import pickle
 import cv2
 import numpy as np
 import torch
-from torch import randint
 
 from dan.manager.dataset import DatasetManager, GenericDataset, apply_preprocessing
-from dan.utils import pad_image, pad_images, pad_sequences_1D, token_to_ind
+from dan.utils import pad_images, pad_sequences_1D, token_to_ind
 
 
 class OCRDatasetManager(DatasetManager):
@@ -47,20 +46,6 @@ class OCRDatasetManager(DatasetManager):
         dataset.charset = self.charset
         dataset.tokens = self.tokens
         dataset.convert_labels()
-        if (
-            "padding" in dataset.params["config"]
-            and dataset.params["config"]["padding"]["min_height"] == "max"
-        ):
-            dataset.params["config"]["padding"]["min_height"] = max(
-                [s["img"].shape[0] for s in self.train_dataset.samples]
-            )
-        if (
-            "padding" in dataset.params["config"]
-            and dataset.params["config"]["padding"]["min_width"] == "max"
-        ):
-            dataset.params["config"]["padding"]["min_width"] = max(
-                [s["img"].shape[1] for s in self.train_dataset.samples]
-            )
 
 
 class OCRDataset(GenericDataset):
@@ -116,34 +101,6 @@ class OCRDataset(GenericDataset):
             [0, sample["img"].shape[0]],
             [0, sample["img"].shape[1]],
         ]
-        # Padding constraints to handle model needs
-        if "padding" in self.params["config"] and self.params["config"]["padding"]:
-            if (
-                self.set_name == "train"
-                or not self.params["config"]["padding"]["train_only"]
-            ):
-                min_pad = self.params["config"]["padding"]["min_pad"]
-                max_pad = self.params["config"]["padding"]["max_pad"]
-                pad_width = (
-                    randint(min_pad, max_pad, (1,))
-                    if min_pad is not None and max_pad is not None
-                    else None
-                )
-                pad_height = (
-                    randint(min_pad, max_pad, (1,))
-                    if min_pad is not None and max_pad is not None
-                    else None
-                )
-
-                sample["img"], sample["img_position"] = pad_image(
-                    sample["img"],
-                    new_width=self.params["config"]["padding"]["min_width"],
-                    new_height=self.params["config"]["padding"]["min_height"],
-                    pad_width=pad_width,
-                    pad_height=pad_height,
-                    padding_mode=self.params["config"]["padding"]["mode"],
-                    return_position=True,
-                )
         return sample
 
     def convert_labels(self):
@@ -178,11 +135,8 @@ class OCRCollateFunction:
         labels = pad_sequences_1D(labels, padding_value=self.label_padding_value)
         labels = torch.tensor(labels).long()
 
-        padding_mode = (
-            self.config["padding_mode"] if "padding_mode" in self.config else "br"
-        )
         imgs = [batch_data[i]["img"] for i in range(len(batch_data))]
-        imgs = pad_images(imgs, padding_mode=padding_mode)
+        imgs = pad_images(imgs)
         imgs = torch.tensor(imgs).float().permute(0, 3, 1, 2)
 
         formatted_batch_data = {