[Target Detection] Teaching uses Colab to train and test YOLO-NAS (first release)

Log in to Google Colab, create a new document, change the running type to GPU

View allocated GPU (typically Tesla T4 16GB)

!nvidia-smi

Install the python library

%?pture
!pip install super-gradients==3.1.0
!pip install imutils
!pip install roboflow
!pip install pytube --upgrade

call the API

from super_gradients.training import Trainer
from super_gradients.training import dataloaders
from super_gradients.training.dataloaders.dataloaders import coco_detection_yolo_format_train, coco_detection_yolo_format_val
from IPython.display import clear_output
from super_gradients.training.losses import PPYoloELosses
from super_gradients.training.metrics import DetectionMetrics_050
from super_gradients.training.models.detection_models.pp_yolo_e import PPYoloEPostPredictionCallback
from super_gradients. training import models

Set training experiment information

CHECKPOINT_DIR = 'checkpoints2'
trainer = Trainer(experiment_name='ppe_yolonas_run2', ckpt_root_dir=CHECKPOINT_DIR)

Download the public dataset

!pip install roboflow
from roboflow import Roboflow
rf = Roboflow(api_key="4v0m0Q44ghy9SpeYcnuN")
project = rf.workspace("objet-detect-yolov5").project("eep_detection-u9bbd")
dataset = project.version(1).download("yolov5")

Set dataset information

dataset_params = {
    'data_dir': '/content/EEP_Detection-1',
    'train_images_dir': 'train/images',
    'train_labels_dir': 'train/labels',
    'val_images_dir': 'valid/images',
    'val_labels_dir': 'valid/labels',
    'test_images_dir': 'test/images',
    'test_labels_dir': 'test/labels',
    'classes': ['Protective Helmet', 'Shield', 'Jacket', 'Dust Mask', 'Eye Wear', 'Glove', 'Protective Boots']
}

Data loader

train_data = coco_detection_yolo_format_train(
    dataset_params={
        'data_dir': dataset_params['data_dir'],
        'images_dir': dataset_params['train_images_dir'],
        'labels_dir': dataset_params['train_labels_dir'],
        'classes': dataset_params['classes']
    },
    dataloader_params={
        'batch_size': 16,
        'num_workers': 2
    }
)

val_data = coco_detection_yolo_format_val(
    dataset_params={
        'data_dir': dataset_params['data_dir'],
        'images_dir': dataset_params['val_images_dir'],
        'labels_dir': dataset_params['val_labels_dir'],
        'classes': dataset_params['classes']
    },
    dataloader_params={
        'batch_size': 16,
        'num_workers': 2
    }
)

test_data = coco_detection_yolo_format_val(
    dataset_params={
        'data_dir': dataset_params['data_dir'],
        'images_dir': dataset_params['test_images_dir'],
        'labels_dir': dataset_params['test_labels_dir'],
        'classes': dataset_params['classes']
    },
    dataloader_params={
        'batch_size': 16,
        'num_workers': 2
    }
)

clear_output()

data conversion

train_data.dataset.transforms

train_data.dataset.dataset_params['transforms'][1]

train_data.dataset.dataset_params['transforms'][1]['DetectionRandomAffine']['degrees'] = 10.42

data visualization

train_data.dataset.plot()

Set up the model (eg YOLO-NAS S)

model = models.get('yolo_nas_s',
                   num_classes=len(dataset_params['classes']),
                   pretrained_weights="coco"
                   )

Set training parameters

train_params = {
    # ENABLING SILENT MODE
    'silent_mode': True,
    "average_best_models": True,
    "warmup_mode": "linear_epoch_step",
    "warmup_initial_lr": 1e-6,
    "lr_warmup_epochs": 3,
    "initial_lr": 5e-4,
    "lr_mode": "cosine",
    "cosine_final_lr_ratio": 0.1,
    "optimizer": "Adam",
    "optimizer_params": {"weight_decay": 0.0001},
    "zero_weight_decay_on_bias_and_bn": True,
    "ema": True,
    "ema_params": {"decay": 0.9, "decay_type": "threshold"},
    "max_epochs": 15,
    "mixed_precision": True,
    "loss": PPYoloELoss(
        use_static_assigner=False,
        num_classes=len(dataset_params['classes']),
        reg_max=16
    ),
    "valid_metrics_list": [
        DetectionMetrics_050(
            score_thres=0.1,
            top_k_predictions=300,
            num_cls=len(dataset_params['classes']),
            normalize_targets=True,
            post_prediction_callback = PPYoloEPostPredictionCallback(
                score_threshold=0.01,
                nms_top_k=1000,
                max_predictions=300,
                nms_threshold=0.7
            )
        )
    ],
    "metric_to_watch": '[email protected]'
}

Download example videos for post-training inference

!gdown "https://drive.google.com/uc?id=1crFwrpMF1OlaJ0ZCZjBNRo9llLEVR8VQ & amp;confirm=t"
!gdown "https://drive.google.com/uc?id=1cTIBNQ1R_7JAOURVv9cJ6P935ym_IkZ0 & amp;confirm=t"
!gdown "https://drive.google.com/uc?id=1256pNK0nQnEDT6FRLQAraTRkOY7BSprq & amp;confirm=t"
!gdown "https://drive.google.com/uc?id=15D71z_g8uxZfXSx2ya3sy4n2-eg53meH & amp;confirm=t"
!gdown "https://drive.google.com/uc?id=1iYW9ZAsYAaHkWZhFVwQh_ch41TMt30-Q & amp;confirm=t"

model training

trainer. train(model=model,
              training_params=train_params,
              train_loader = train_data,
              valid_loader=val_data)

save the best model

best_model = models.get('yolo_nas_s',
                        num_classes=len(dataset_params['classes']),
                        checkpoint_path="checkpoints2/ppe_yolonas_run2/ckpt_best.pth")

Test Results

trainer.test(model=best_model,
            test_loader=test_data,
            test_metrics_list=DetectionMetrics_050(score_thres=0.1,
            top_k_predictions=300,
            num_cls=len(dataset_params['classes']),
            normalize_targets=True,
            post_prediction_callback=PPYoloEPostPredictionCallback(score_threshold=0.01,
            nms_top_k=1000,
            max_predictions=300,
            nms_threshold=0.7)
            ))

video reasoning

input_video_path = f"/content/demonew2.mp4"
output_video_path = "detections.mp4"
import torch
device = 'cuda' if torch.cuda.is_available() else "cpu"
best_model.to(device).predict(input_video_path).save(output_video_path)