Mirrors/blakeblackshear.frigate
1
0
Fork 0
mirror of https://github.com/blakeblackshear/frigate.git synced 2025-08-27 13:47:50 +02:00
Code Issues Projects Releases Wiki Activity

Upgrade PaddleOCR models to v4 (rec) and v5 (det) (#18505)

Browse Source 
The PP_OCRv5 text detection models have greatly improved over v3. The v5 recognition model makes improvements to challenging handwriting and uncommon characters, which are not necessary for LPR, so using v4 seemed like a better choice to continue to keep inference time as low as possible. Also included is the full dictionary for Chinese character support.
This commit is contained in:
Josh Hawkins 2025-06-01 15:21:12 -05:00 committed by GitHub
parent 568e620963
commit 645868e099
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 126 additions and 111 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

28
frigate/data_processing/common/license_plate/mixin.py
View File

@ -22,7 +22,7 @@ from frigate.comms.event_metadata_updater import (
EventMetadataPublisher, EventMetadataPublisher,
EventMetadataTypeEnum, EventMetadataTypeEnum,
) )
from frigate.const import CLIPS_DIR from frigate.const import CLIPS_DIR, MODEL_CACHE_DIR
from frigate.embeddings.onnx.lpr_embedding import LPR_EMBEDDING_SIZE from frigate.embeddings.onnx.lpr_embedding import LPR_EMBEDDING_SIZE
from frigate.types import TrackedObjectUpdateTypesEnum from frigate.types import TrackedObjectUpdateTypesEnum
from frigate.util.builtin import EventsPerSecond, InferenceSpeed from frigate.util.builtin import EventsPerSecond, InferenceSpeed
@ -43,7 +43,11 @@ class LicensePlateProcessingMixin:
self.plates_det_second = EventsPerSecond() self.plates_det_second = EventsPerSecond()
self.plates_det_second.start() self.plates_det_second.start()
self.event_metadata_publisher = EventMetadataPublisher() self.event_metadata_publisher = EventMetadataPublisher()
self.ctc_decoder = CTCDecoder() self.ctc_decoder = CTCDecoder(
character_dict_path=os.path.join(
MODEL_CACHE_DIR, "paddleocr-onnx", "ppocr_keys_v1.txt"
)
)
self.batch_size = 6 self.batch_size = 6
# Detection specific parameters # Detection specific parameters
@ -1600,14 +1604,21 @@ class CTCDecoder:
for each decoded character sequence. for each decoded character sequence.
""" """
def __init__(self): def __init__(self, character_dict_path=None):
""" """
Initialize the CTCDecoder with a list of characters and a character map. Initializes the CTCDecoder.
:param character_dict_path: Path to the character dictionary file.
The character set includes digits, letters, special characters, and a "blank" token If None, a default (English-focused) list is used.
(used by the CTC model for decoding purposes). A character map is created to map For Chinese models, this should point to the correct
indices to characters. character dictionary file provided with the model.
""" """
self.characters = []
if character_dict_path and os.path.exists(character_dict_path):
with open(character_dict_path, "r", encoding="utf-8") as f:
self.characters = ["blank"] + [
line.strip() for line in f if line.strip()
]
else:
self.characters = [ self.characters = [
"blank", "blank",
"0", "0",
@ -1707,6 +1718,7 @@ class CTCDecoder:
" ", " ",
" ", " ",
] ]
self.char_map = {i: char for i, char in enumerate(self.characters)} self.char_map = {i: char for i, char in enumerate(self.characters)}
def __call__( def __call__(

11
frigate/embeddings/onnx/lpr_embedding.py
View File

@ -32,13 +32,15 @@ class PaddleOCRDetection(BaseEmbedding):
device: str = "AUTO", device: str = "AUTO",
): ):
model_file = ( model_file = (
"detection-large.onnx" if model_size == "large" else "detection-small.onnx" "detection_v5-large.onnx"
if model_size == "large"
else "detection_v5-small.onnx"
) )
super().__init__( super().__init__(
model_name="paddleocr-onnx", model_name="paddleocr-onnx",
model_file=model_file, model_file=model_file,
download_urls={ download_urls={
model_file: f"https://github.com/hawkeye217/paddleocr-onnx/raw/refs/heads/master/models/{model_file}" model_file: f"https://github.com/hawkeye217/paddleocr-onnx/raw/refs/heads/master/models/v5/{model_file}"
}, },
) )
self.requestor = requestor self.requestor = requestor
@ -156,9 +158,10 @@ class PaddleOCRRecognition(BaseEmbedding):
): ):
super().__init__( super().__init__(
model_name="paddleocr-onnx", model_name="paddleocr-onnx",
model_file="recognition.onnx", model_file="recognition_v4.onnx",
download_urls={ download_urls={
"recognition.onnx": "https://github.com/hawkeye217/paddleocr-onnx/raw/refs/heads/master/models/recognition.onnx" "recognition_v4.onnx": "https://github.com/hawkeye217/paddleocr-onnx/raw/refs/heads/master/models/v4/recognition_v4.onnx",
"ppocr_keys_v1.txt": "https://github.com/hawkeye217/paddleocr-onnx/raw/refs/heads/master/models/v4/ppocr_keys_v1.txt",
}, },
) )
self.requestor = requestor self.requestor = requestor