YOLO-NAS architecture is out! The brand new YOLO-NAS delivers state-of-the-art performance with the unparalleled accuracy-speed performance, outperforming other models corresponding to YOLOv5, YOLOv6, YOLOv7 and YOLOv8.

We’ll understand How one can train YOLO-NAS on Custom Dataset.I’m gonna use Roboflow Dataset.

Dataset Link — https://universe.roboflow.com/object-detect-dmjpt/waste-9cbfm

We’ll use google colab for training our dataset

YOLO-NAS Starter Notebook —https://colab.research.google.com/drive/1q0RmeVRzLwRXW-h9dPFSOchwJkThUy6d?pli=1

Step 1.Installation of Required Library Might take jiffy.

Step 2.Restart Runtime before moving forward Necessary Step

Step 3. Download The pretrained weights — yolo_nas_s or depend upon your requirements.

Run the following code cell remember to say the right model = yolo_nas_s in my case.

Run this above cell.

Step 4. High quality Tuning YOLO-NAS on custom Dataset

Check the directory is created

Step 5. Let take the dataset from roboflow I actually have already mention dataset link on top.

Run this above cell code for importing the required modules provided by SuperGradients DataLoaders.

Most Necessary step while loading the dataset for training.

Change the batch size if required I actually have change it 4.

Run this all below shown code cells.

Step 6. As I already mentioned I’m using yolo_nas_s model for this training.

Defining the Metrics and Training Parameters

Integrations with experiment monitoring tools.SuperGradients has native integrations with Tensorboard, Weights and Biases, ClearML, and DagsHub.

Step 7. Let start the training.

Output of the training we appear like this it would take time to display the output.

Finally Training is over.You’re going to get your best and last weights within the checkpoint folder.

Step 8. Evaluating the perfect trained model on the test set.

Step 9. Let do detection using the perfect model.

A. FOR IMAGES.

Run this below code cell use any image give proper path and it display the anticipated result

B. FOR VIDEOS.

Output shall be stored in content folder in .mp4 format.

Thank You

I actually have made tutorial video also on this topic You may checkout this on my YouTube Channel.