YOLO26

A wrapper implementation for YOLOv26 (Ultralytics YOLO v8.4.0+) object detection within the ECOS AI framework. This module provides an easy-to-use interface for training, inference, and visualization of object detection tasks, segmentation,...

Overview

The YOLO26 module extends the BaseModel class and integrates the latest Ultralytics YOLO implementation for object detection tasks. It supports automatic model downloading, training, prediction, and result visualization with customizable parameters.

Installation

Please install ecos_core

Requirements

pip install -r requirements.txt

Usage

Configuration

Configuration is managed through a JSON file (opt.json). Here's the example structure:

{
    "data": "/path/to/data.yaml",
    "weight": "yolo26s.pt",
    "weights_url": "https://github.com/ultralytics/assets/releases/download/v8.4.0/yolo26s.pt",
    "batch_size": 8,
    "epochs": 1500,
    "device": 0,
    "imgsz": 640,
    "task": "detect",
    "type": "pytorch",
    "conf": 0.25,
    "iou": 0.7,
    "font_scale": 2,
    "text_thickness": 2,
    "padding_rect": 5,
    "visualization": true
}

With: - data: Path to data.yaml file containing dataset configuration - weight: Model weight filename - weights_url: URL to download pre-trained weights - batch_size: Number of samples per training batch - epochs: Number of training epochs - device: Device to use (0,1,2,3 for GPU or "cpu") - imgsz: Target image size for training. - task: Task type. Default: "detect" - type: Framework type - conf: Confidence threshold for predictions - iou: IoU threshold for NMS - font_scale: Font scale for visualization text - text_thichness: Text thickness for visualization - padding_rect: Padding for text background rectangles - visualization: Enable/disable result visualization

Training

Basic Training Example:

import sys
import os
import json
import argparse
sys.path.insert(0, os.path.abspath(
    os.path.join(os.path.dirname(__file__), './../../')
))
from ecos_core.yolo26.yolo26 import Model
from easydict import EasyDict

if __name__ == "__main__":
    parser = argparse.ArgumentParser()
    parser.add_argument('--opt',
                        type=str, default="./opt.json", help='opt path')
    parser.add_argument('--weight-path',
                        type=str, default=None, help='weight path')
    args = parser.parse_args()
    opt = args.opt
    weight_path = args.weight_path
    with open(opt, "r") as f:
        opts = EasyDict(json.load(f))
    model = Model(opts)
    print(weight_path)
    model.build_model(weight_path)
    model.fit()

Training script:

python train.py --opt ./opt.json --weight-path ./yolo26s.pt

Inference

Basic inference example:

import sys
import os
import json
import argparse
sys.path.insert(0, os.path.abspath(
    os.path.join(os.path.dirname(__file__), './../../')
))
from ecos_core.yolo26.yolo26 import Model
from easydict import EasyDict

if __name__ == "__main__":
    parser = argparse.ArgumentParser()
    parser.add_argument('--opt',
                        type=str, default="./opt.json", help='opt path')
    parser.add_argument('--weight-path',
                        type=str, default=None, help='weight path')
    parser.add_argument('--input-path', required=True,
                        type=str, default="./broken/images/test", help='input path')
    args = parser.parse_args()
    opt = args.opt
    weight_path = args.weight_path
    with open(opt, "r") as f:
        opts = EasyDict(json.load(f))
    model = Model(opts)
    print(weight_path)
    model.build_model(weight_path)
    save_path = model.process_predictions(
        net_output=results,
        raw_image_path="path/to/image.jpg",
        raw_image_mat=None,
        image_transform=None,
        save_image_path="output/annotated_image.jpg"
    )

APIs

`Model`

Bases: BaseModel

`build_model(custom_weight=None)`

Build model in YOLO26 model

Parameters:

Name	Type	Description	Default
`custom_weight`	`str`	custom weight path. Defaults to None.	`None`

Returns:

Name	Type	Description
`model`	`Model`	YOLO model

`fit()`

Train the model

Returns:

Name	Type	Description
`results`		training results

`get_transform()`

Get image data transform function for preprocessing

Returns:

Name	Type	Description
`transform`	`func`	transform function. This function will take input as image path and output - raw_image: image numpy mat - image_transform: image tensor (pytorch) Example: torchvision.transforms.transforms: data transforms function

`predict(source=None, classes=None)`

Run inference on images or folders.

Parameters:

Name	Type	Description	Default
`source`	`str`	image path or folder path. Defaults to None.	`None`
`classes`	`list`	list of class ids to filter. Defaults to None.	`None`

Returns:

Name	Type	Description
`results`		prediction results

`process_predictions(net_output, raw_image_path, raw_image_mat, image_transform, save_image_path)`

Process predictions and generate annotated images.

Parameters:

Name	Type	Description	Default
`net_output`		output of detection net	required
`raw_image_path`	`str`	raw image path	required
`raw_image_mat`	`Array`	raw image numpy mat type	required
`image_transform`	`func`	image_transform function	required
`save_image_path`	`str`	save image path	required

Returns:

Name	Type	Description
`save_image_path`		inform that the annotation image has been written successfully.

`reload_param(opt_path='./opt.json')`

Reload parameters

Parameters:

Name	Type	Description	Default
`opt_path`	`str`	Opt path. Defaults to "./opt.json".	`'./opt.json'`

`update_opt(option, output_path='./opt.json')`

Update config

Parameters:

Name	Type	Description	Default
`option`	`Namespace / Dict`	option to update	required
`output_path`	`str`	path to save opt json. Defaults to "./opt.json".	`'./opt.json'`