[refactor] Init

texteller/constants.py

@@ -0,0 +1,33 @@
+# Formula image(grayscale) mean and variance
+IMAGE_MEAN = 0.9545467
+IMAGE_STD = 0.15394445
+
+# Vocabulary size for TexTeller
+VOCAB_SIZE = 15000
+
+# Fixed size for input image for TexTeller
+FIXED_IMG_SIZE = 448
+
+# Image channel for TexTeller
+IMG_CHANNELS = 1  # grayscale image
+
+# Max size of token for embedding
+MAX_TOKEN_SIZE = 1024
+
+# Scaling ratio for random resizing when training
+MAX_RESIZE_RATIO = 1.15
+MIN_RESIZE_RATIO = 0.75
+
+# Minimum height and width for input image for TexTeller
+MIN_HEIGHT = 12
+MIN_WIDTH = 30
+
+LATEX_DET_MODEL_URL = (
+    "https://huggingface.co/TonyLee1256/texteller_det/resolve/main/rtdetr_r50vd_6x_coco.onnx"
+)
+TEXT_REC_MODEL_URL = (
+    "https://huggingface.co/OleehyO/paddleocrv4.onnx/resolve/main/ch_PP-OCRv4_server_rec.onnx"
+)
+TEXT_DET_MODEL_URL = (
+    "https://huggingface.co/OleehyO/paddleocrv4.onnx/resolve/main/ch_PP-OCRv4_det.onnx"
+)