memolib 1.0.2

A comprehensive OCR and computer vision model library with support for SVTRV2, YOLO, EfficientNet, PPLCNet, and more.

MemoLib

A comprehensive OCR and computer vision model library built on PyTorch, providing a unified interface for text recognition, object detection, image segmentation, and image classification.

Features

• Text Recognition - SVTRV2 model with GTC (Generative Text Context) pipeline, SMTR decoder, and RCTC loss
• Object Detection - YOLO family (v5, v8, v11, v26) with detection and segmentation variants
• Image Segmentation - DINO + UperNet semantic segmentation
• Image Classification - EfficientNet, PPLCNet, YOLO classification
• Unified Interface - Consistent API across all models: LoadWeight, Predict, BatchPredict, Train, Export
• Model Export - Export to ONNX, OpenVINO, or TensorRT formats

Installation

pip install memolib

Optional dependencies

# For YOLO models
pip install memolib[yolo]

# For PPLCNet models
pip install memolib[pplcnet]

# For development
pip install memolib[dev]

Quick Start

Text Recognition with SVTRV2

from MemoLib import SVTRV2, TrainingConfig

# Load a pre-trained recognition model
model = SVTRV2()
model.LoadWeight("path/to/weights.pth")

# Single prediction
result = model.Predict(image)

# Batch prediction
results = model.BatchPredict([img1, img2, img3])

Object Detection with YOLO

from MemoLib import Yolo, TrainingConfig
from MemoLib.Model.BaseModel import eYoloDetectionModel

# Load a YOLO detection model
model = Yolo()
model.LoadWeight(eYoloDetectionModel.YoloV8n)

# Single prediction
result = model.Predict(image)

Semantic Segmentation with DINO UperNet

from MemoLib.Model.DinoUperNet import DinoUperNet, TrainingConfig

model = DinoUperNet()
model.LoadWeight("path/to/weights.pth")
result = model.Predict(image)

Model Export

from MemoLib.Model.BaseModel import eModelExportType

# Export to ONNX
model.Export("model.onnx", eModelExportType.ONNX)

# Export to OpenVINO
model.Export("model.xml", eModelExportType.OpenVINO)

# Export to TensorRT
model.Export("model.engine", eModelExportType.TensorRT)

Project Structure

MemoLib/
├── Model/
│   ├── BaseModel/       # Base interfaces and model type enums
│   │   ├── IModel.py    # Abstract model interface
│   │   └── eModelBase.py # Model type enumerations
│   ├── STVRV2/          # SVTRV2 OCR recognition model
│   ├── YOLO/            # YOLO detection & classification
│   ├── DinoUperNet/     # DINO + UperNet segmentation
│   ├── Efficientnet/    # EfficientNet classification
│   └── PPLCNet/         # PPLCNet classification

Available Model Types

Category	Models
Recognition	SVTRV2 (GTC/SMTR/RCTC)
Detection	YOLOv5, YOLOv8, YOLO11, YOLO26, RF-DETR
Segmentation	YOLO Segment, DINO UperNet
Classification	EfficientNet (B0-B7, V2), PPLCNet, YOLO Class, RF-DETR Class
Anomaly Detection	DInomaly, INFormerly

Requirements

• Python 3.9+
• PyTorch >= 1.10.0
• torchvision >= 0.11.0
• numpy >= 1.21.0
• opencv-python >= 4.5.0

License

MIT License