Get Started

Import the Dataset and Model

Copy the tutorial inputs into your Project and confirm their task settings.

Import copies each external source into the Project as an independently owned, content-identified Material. The original download remains separate.

Import the Dataset

  1. Open Datasets > Import.
  2. Select the extracted Dataset folder.
  3. Confirm YOLO Object Detection as the detected format.
  4. Confirm the Dataset split and class information published on the download page.
  5. Start the import and wait for it to complete.

Import the complete folder as Helmet v2 full dataset: 2,335 images (Train 2,043 / Validation 195 / Test 97). Random sampling happens later on the Prepare Prediction Set page, not during import.

Imported Helmet v2 full dataset Dataset with annotated samples and split and label controls.

Watch: Inspect detection samplesToggle box annotations and review the imported dataset format.Silent demo · Subtitles follow page language · Full screen for detail

Import may continue while you visit another screen. Use the Runtime monitor or the running item in the Datasets list to return to its progress. If it fails, read the retained diagnostic before starting a fresh import.

Import the Model

  1. Open Assets > Local Models > Import.
  2. Select the downloaded .pt file.
  3. Confirm the Model name, YOLO backend, and Object Detection task.
  4. Enter the exact class labels and inference settings published with the Model.
  5. Save the Model. Knotest verifies it when preparing Prediction Generation.

The documented demo was verified with helmet-object-detection-best.pt:

SettingValue
Backend / taskYOLO / Object Detection
Class order0: helmet
Input dimensions640 × 640
Minimum object confidence0.25
Duplicate-box overlap (NMS IoU)0.5
Class-agnostic NMSOff
Maximum detections300
Tutorial splitValidation: random 100 of 195, seed 42

Confidence, overlap, and split are reviewed in Prediction Generation's Run options. The Model detail below shows class order and input dimensions.

Helmet YOLO26 detector Model settings with the helmet class and 640 by 640 input dimensions.

Watch: Review model settingsOpen the imported detector and inspect its class order and input dimensions.Silent demo · Subtitles follow page language · Full screen for detail

Do not change the label order to make names look more natural. Dataset and Model class alignment is part of the Evaluation input contract.

Continue to Generate a Prediction Set.

Create a Project

Create the visible folder that will own the tutorial evaluation record.

Generate a Prediction Set

Generate clean and Fog-transformed predictions before scoring them.

On this page

Import the DatasetImport the Model