Its the first task of TeknoFest 2021 autonomous underwater vehicles category which is a technology competition in Turkey.
Dataset
Our vehicle can detects a custom object in underwater. I created my own dataset by taking pictures of our custom gate. I use 25 images for training and 9 images for test. I labelled the images by using LabelImg.
LabelImg: https://github.com/tzutalin/labelImg.git
I upload my dataset to kaggle you can see here https://www.kaggle.com/sedanurkrc/underwater-gate-dataset
Here is some pictures of the dataset
Training
I use ssd_mobilenet_v2 pretrainede models weights. And in 2000 epochs it gets 0.18 loss.
Test
