Hamoye : Team SMOTE

Object detection using TensorFlow and YOLOv5

Approach:

Dataset used: African Wildlife

The original dataset was YOLO specific so our team split into two parts:

1. Team Tensorflow

Deployed Model Link

WorkFlow:

• Data Collection
• Data Cleaning
• Data Augmentation
• Modelling
• Evaluation
• Deployment

For the tensorflow models, we collected animal images of animlas through data scrapping as our YOLO dataset had images with multiple animals in them reducing the accuracy of the model.

The team used the following models:

• EfficientNet
• Inception V3
• MobileNet V2
• NASNetLarge
• Resnet152
• Resnet50
• VGG16

And finally the EfficientNetB7 model was selected for deployment.

2. Team YOLO

Deployed Model Link

For the YOLO model, we used the original kaggle dataset and used YOLOv5x6 for the final modelling.

WorkFlow:

• Modelling
• Evaluation
• Deployment

YOLO Video Tests

YOLO.Test.1.mp4

YOLO.Test.2.mp4

YOLO.Test.3.mp4