Object Segmentation

Welcome to my Object Segmentation project, where I utilized machine learning to accurately detect and segment oranges and limes in images. This project aims to demonstrate the power of object segmentation using machine learning and how it can be applied to improve image analysis.

Object Segmentation

Object segmentation is the process of identifying and isolating specific objects within an image. For this project, I utilized MobileNetV2 to accurately detect and segment oranges and limes in images.

Dataset

To train the models, I used a variety of datasets, including ImageNet and Fruits 360, to obtain the necessary images of oranges and limes. Additionally, I utilized the COCO dataset, which contains 80 categories of objects, to train the models to recognize oranges in images.

Labeling

For labeling, I used various tools, including LabelBox, Label Studio, and Diffgram. I found that LabelBox was the best tool for uploading large datasets and had better iPad usability. I encountered challenges, such as dealing with various textured limes and reviewing submitted annotations, but found that using MAL was helpful in speeding up the labeling process.

Training

The models were trained on a custom additional training set that I created using various fruit datasets. Additionally, I employed Model-assisted labeling (MAL) to speed up the labeling process and reduce the time and effort required to train the models. I utilized AWS EC2 p3.8xlarge, which is equipped with four Tesla V100s of 64GB GPU memory, to perform the training.

Orange and Lime Segmentation

For orange segmentation, I utilized MobileNetV2 and trained the model on the COCO dataset for oranges, achieving an accuracy of 94% on 1699 images. The segmented oranges are colored in a rainbow pattern, making them easy to distinguish from other objects. I also created a lime segmentation model using MobileNetV2, which showed 70% accuracy when trained on ~200 lime images. This model will be used for MAL in LabelBox.

Conclusion

In summary, my Object Segmentation project demonstrates the power of machine learning for accurately detecting and segmenting oranges and limes in images. Through the use of various tools, including MobileNetV2, and datasets, including COCO, ImageNet, and Fruits 360, I was able to train accurate models for object segmentation. I also learned about the importance of labeling conventions and conversions, the benefits of open-source datasets, and the potential for labeling group projects using labeling platforms with AI labeling features. I hope this project inspires further research and applications in the field of computer vision.

2022