ULTIMA-YOLOv9

About ULTIMA-YOLO models

<div style="display: flex;"> <img src="./0th.png" style="width: auto; height: auto; margin-right: 10px;"> </div> <div style="display: flex;"> <img src="./1st.jpg" style="width: auto; height: auto; margin-right: 10px;"> </div> <div style="display: flex;"> <img src="./2nd.png" style="width: auto; height: auto; margin-right: 10px;"> </div> <div style="display: flex;"> <img src="./3rd.jpg" style="width: auto; height: auto; margin-right: 10px;"> </div>

This is a part of ULTIMA project.

ULTIMA is Uma Musume Labeled Text-Image Multimodal Alignment.

ULTIMA-YOLOv9 model is a facial detection model for Uma Musumes in illustrations and based on yolov9-e and ULTIMA-YOLO dataset

This is the model repository for ULTIMA-YOLOv9, containing the following checkpoints:

• YOLO9-E

How to Use

Clone YOLOv9 repository.

git clone https://github.com/WongKinYiu/yolov9.git
cd yolov9

Download the weights using hf_hub_download and use the loading function in helpers of YOLOv9.

from huggingface_hub import hf_hub_download 
hf_hub_download("UmaDiffusion/ULTIMA-YOLOv9", filename="ultima_yolov9-e.pt", local_dir="./")

Load the model.

# make sure you have the following dependencies
import torch
import numpy as np
from models.common import DetectMultiBackend
from utils.general import non_max_suppression, scale_boxes
from utils.torch_utils import select_device, smart_inference_mode
from utils.augmentations import letterbox
import PIL.Image

@smart_inference_mode()
def predict(image_path, weights='ultima_yolov9-e.pt', imgsz=640, conf_thres=0.1, iou_thres=0.45):
    # Initialize
    device = select_device('0')
    model = DetectMultiBackend(weights=weights, device=device, fp16=False)
    stride, names, pt = model.stride, model.names, model.pt

    # Load image
    image = np.array(PIL.Image.open(image_path).convert("RGB"))
    img = letterbox(image, imgsz, stride=stride, auto=True)[0]
    img = img.transpose(2, 0, 1)
    img = np.ascontiguousarray(img)
    img = torch.from_numpy(img).to(device).float()
    img /= 255.0
    if img.ndimension() == 3:
        img = img.unsqueeze(0)

    # Inference
    pred = model(img, augment=False, visualize=False)

    # Apply NMS
    pred = non_max_suppression(pred[0][0], conf_thres, iou_thres, classes=None, max_det=1000)

or use detect.py in yolov9 repo.

python ./detect.py --source [image_path] --device 0 --img 1280 --weights './ultima_yolov9-e.pt' --name ultima_yolov9_1280_detect

Training Infomation

• Batch Size: 32
• Resolution: 640
• Epochs: 300, chosen best mAP
• GPU: 1x A6000 48GB
• Dataset: ULTIMA-YOLO

Statistics

• Train: 3,991 images
• Val: 399 images