jupyter
accelerator colab kernelspec language_info nbformat nbformat_minor GPU gpuType include_colab_link T4 true display_name name Python 3 python3 name python 4 0

How to Train YOLO11 Object Detection Model on a Custom Dataset :::

outputId="0f37daf0-fff1-45e7-a9e6-9464c5596f22"}

!nvidia-smi

Step 01 # Install the Ultralytics Package :::

!pip install ultralytics

Step 02 # Import All the Requried Libraries :::

import ultralytics
ultralytics.checks()

from ultralytics import YOLO
from IPython.display import Image

Step # 03 Download Dataset from Roboflow :::

!pip install roboflow

from roboflow import Roboflow
rf = Roboflow(api_key="____________________")
project = rf.workspace("test-tnuvx").project("extracting-facial-features-ykifq-bjqdx")
version = project.version(3)
dataset = version.download("yolov11")

dataset.location

Step # 04 Train YOLO11 Model on a Custom Dataset :::

!yolo task=detect mode=train data={dataset.location}/data.yaml model="yolo11n.pt" epochs=40 imgsz=640

Step # 05 Examine Training Results :::

Image("/content/runs/detect/train4/P_curve.png", width=600)

::: {.output .execute_result execution_count="32"}

Image("/content/runs/detect/train4/R_curve.png", width=600)

::: {.output .execute_result execution_count="33"}

::: {.cell .code colab="{"base_uri":"https://localhost:8080/\",\"height\":467}" id="vHVXACym1VZk" outputId="841a8187-0b0f-432b-85f9-ce236f49403c"}

Image("/content/runs/detect/train4/confusion_matrix.png", width=600)

::: {.output .execute_result execution_count="34"} {width="600"} ::: :::

::: {.cell .code colab="{"base_uri":"https://localhost:8080/\",\"height\":467}" id="GMMRL5LT1b0L" outputId="5acb0b5d-46ac-4b0f-d522-7209e07e8c3a"}

Image("/content/runs/detect/train4/confusion_matrix_normalized.png", width=600)

::: {.output .execute_result execution_count="35"} {width="600"} ::: :::

::: {.cell .code colab="{"base_uri":"https://localhost:8080/\",\"height\":617}" id="bffrCmrH1izb" outputId="08b7fd42-1734-4bd3-e2c1-1d5e5c23e85f"}

Image("/content/runs/detect/train4/labels.jpg", width=600)

::: {.output .execute_result execution_count="36"} {width="600"} ::: :::

::: {.cell .code colab="{"base_uri":"https://localhost:8080/\",\"height\":317}" id="PeSiFrwP1tbF" outputId="be3db45d-976d-4fc4-efb9-41a97523ac90"}

Image("/content/runs/detect/train4/results.png", width=600)

::: {.output .execute_result execution_count="37"} {width="600"} ::: :::

::: {.cell .code colab="{"base_uri":"https://localhost:8080/\",\"height\":617}" id="RacyVWK3103E" outputId="b46510b0-a37f-4547-ec7d-48cd39e2953f"}

Image("/content/runs/detect/train4/train_batch0.jpg", width=600)

::: {.output .execute_result execution_count="38"} {width="600"} ::: :::

::: {.cell .code colab="{"base_uri":"https://localhost:8080/\",\"height\":617}" id="idaZDidT10_L" outputId="bff75243-089a-41b7-d04f-ac8d147fca14"}

Image("/content/runs/detect/train4/val_batch0_pred.jpg", width=600)

::: {.output .execute_result execution_count="39"} {width="600"} ::: :::

::: {.cell .code colab="{"base_uri":"https://localhost:8080/\",\"height\":757}" id="VPEj-hRA11Pn" outputId="de5aae30-862b-4d22-ff03-ab3bddc07f1f"}

Image("/content/runs/detect/train4/val_batch1_pred.jpg", width=600)

::: {.output .execute_result execution_count="40"}

Image("/content/runs/detect/train4/val_batch2_pred.jpg", width=600)

::: {.output .execute_result execution_count="41"}

::: {.cell .markdown id="T8eXR6l03d-Q"} Step # 06 Download the Model Weights from the Google Drive :::

::: {.cell .markdown id="alfU2d0z41X6"} Step # 07 Validate Fine-Tuned Model :::

!yolo task=detect mode=val model="/content/runs/detect/train4/weights/best.pt" data={dataset.location}/data.yaml

::: {.output .stream .stdout} Ultralytics 8.3.73 🚀 Python-3.11.11 torch-2.5.1+cu124 CUDA:0 (Tesla T4, 15095MiB) YOLO11n summary (fused): 238 layers, 2,583,907 parameters, 0 gradients, 6.3 GFLOPs val: Scanning /content/Extracting-Facial-Features-3/valid/labels.cache... 153 images, 0 backgrounds, 0 corrupt: 100% 153/153 [00:00<?, ?it/s] Class Images Instances Box(P R mAP50 mAP50-95): 100% 10/10 [00:03<00:00, 2.95it/s] all 153 765 0.9 0.789 0.842 0.739 eye 153 153 1 0.98 0.995 0.84 eyebrow 153 153 1 0.981 0.995 0.822 lip 153 153 1 0.993 0.995 0.896 mustache-beard 1 1 0.701 1 0.995 0.895 nose 153 153 1 1 0.995 0.892 sharp-jawline 27 27 0.605 0.624 0.638 0.574 slight-double-chin 5 5 1 0 0.196 0.178 soft-jawline 120 120 0.898 0.734 0.928 0.815 Speed: 2.1ms preprocess, 7.1ms inference, 0.0ms loss, 2.3ms postprocess per image Results saved to runs/detect/val 💡 Learn more at https://docs.ultralytics.com/modes/val ::: :::

::: {.cell .markdown id="OOdt0GSA5ntm"} Step # 08 Inference with Custom Model on Images :::

::: {.cell .code colab="{"base_uri":"https://localhost:8080/\"}" collapsed="true" id="E1dU-fiz3sUr" outputId="8cb86c21-c666-450f-e82f-5f96d0ff8298"}

!yolo task=detect mode=predict model="/content/runs/detect/train4/weights/best.pt" conf=0.25 source={dataset.location}/test/images save=True

import glob
import os
from IPython.display import Image as IPyImage, display

latest_folder = max(glob.glob('/content/runs/detect/predict/'), key=os.path.getmtime)
for img in glob.glob(f'{latest_folder}/*.jpg')[1:4]:
    display(IPyImage(filename=img, width=600))
    print("\n")

::: {.output .display_data} {width="600"} :::

::: {.output .stream .stdout} :::

::: {.output .display_data} {width="600"} :::

::: {.output .stream .stdout} :::

::: {.output .display_data} {width="600"} :::

::: {.output .stream .stdout} ::: :::

::: {.cell .code id="uyJMJ5-m7rAt"}

# delete folder and its content
#import shutil
#shutil.rmtree('/content/runs/detect/predict3')

:::

!yolo task=detect mode=predict model="/content/runs/detect/train4/weights/best.pt" conf=0.25 source='' save=True

Step # 08 Inference with Custom Model on Videos :::

!yolo task=detect mode=predict model="/content/runs/detect/train4/weights/best.pt" conf=0.25 source="/content/amazon.mp4" save=True

!rm '/content/result_compressed.mp4'

from IPython.display import HTML
from base64 import b64encode
import os

# Input video path
save_path = '/content/runs/detect/predict4/amazon.avi'

# Compressed video path
compressed_path = "/content/result_compressed.mp4"

os.system(f"ffmpeg -i {save_path} -vcodec libx264 {compressed_path}")

# Show video
mp4 = open(compressed_path,'rb').read()
data_url = "data:video/mp4;base64," + b64encode(mp4).decode()
HTML("""
<video width=400 controls>
      <source src="%s" type="video/mp4">
</video>
""" % data_url)