Learn how to use CNNs: train from scratch, finetune a pretrained model, use a pre-trained model as it is.
The target task to be performed was Image Classification.
The Inaturalist 12K Dataset having 10 classes was used.
Train data (10K Images) was split into (90% Train -10% val).
This implementation is based on Tensorflow.
In part A, we train a CNN model from scratch.
Part A readme describes how to use the code to define a CNN model and train it from scratch.
In part B, we fine-tune models pretrained on Imagenet on the Inaturalist Dataset. Particularly, we experimented with InceptionV3, Xception ResNet50, InceptionResNetV2 MobileNetV2.
Part B readme describes how to use the code to fine-tune pretrained models on the Inaturalist Dataset.
Application chosen: Pothole Detection using Yolo V4. We have used the pre-trained weights of YoloV4 by Alexey Bochkovskiy et al. In part C, we fine-tune YoloV4 to detect road potholes in videos.