https://github.com/ashrockzzz2003/image_segmentation_oxford_pets_dataset

Image Segmentation using Oxford Pets Dataset with the goal to improve animal tranquilizer aiming system.
https://github.com/ashrockzzz2003/image_segmentation_oxford_pets_dataset

apple-gpu fpn metal mts neural-networks opencv resnet34 segmentation tensorflow torch

Last synced: 7 days ago
JSON representation

Image Segmentation using Oxford Pets Dataset with the goal to improve animal tranquilizer aiming system.

• Host: GitHub
• URL: https://github.com/ashrockzzz2003/image_segmentation_oxford_pets_dataset
• Owner: Ashrockzzz2003
• Created: 2024-11-02T06:59:13.000Z (2 months ago)
• Default Branch: main
• Last Pushed: 2024-12-20T05:20:05.000Z (14 days ago)
• Last Synced: 2024-12-20T06:23:49.508Z (14 days ago)
• Topics: apple-gpu, fpn, metal, mts, neural-networks, opencv, resnet34, segmentation, tensorflow, torch
• Language: Jupyter Notebook
• Homepage:
• Size: 10.1 MB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# Optimize Animal Tranquilizer Aiming System

> [!Important]

> What's interesting is that the training data is fully cats and dogs, but the model works with new animals too! 

> The model is trained on the Oxford IIIT Pets dataset, which contains 37 classes of pets. 

> The model can be used for any animal.

https://github.com/user-attachments/assets/43c4bf6d-5754-47bc-aa2c-c2daaedee4ef

Term project for the course: `Neural Networks and Deep Learning`. This project emphasizes improving the tracking of animals in real-time to ensure that the tranquilizer accurately aims at the target.

The dataset used is [Oxford IIIT Pets dataset](https://www.robots.ox.ac.uk/~vgg/data/pets/). The dataset itself contains the respective maskings.

![image](https://github.com/user-attachments/assets/a3c593a9-3169-48a2-a2fd-41cc6d564f4e)

We have developed a model to track the motion of the animal by masking the animal to represent the trace of the tracking achieved by our model.

We have worked with three models, namely:

1. Simple Unet `tensorflow` `trained in Google Colab TPU`

2. Unet CNN `tensorflow` `trained in Google Colab TPU`

3. FPN with ResNet34 `pytorch` `trained in Apple M2 Pro with Metal GPU 19 cores.`

## Evaluation Plots

### Simple Unet  

    

![image](https://github.com/user-attachments/assets/451a21b2-0467-4e9c-8224-b8173bda2f65)

### Unet CNN  

![image](https://github.com/user-attachments/assets/59e5c650-0af2-4eb3-9400-b200c5dd00bf)

### FPN with ResNet34  

![image](https://github.com/user-attachments/assets/0797877c-1fad-4ce7-bc32-1226ec20c6b2)

## Model Comparison

![image](https://github.com/user-attachments/assets/a81e1233-6c19-4bc1-8971-7237bae3fd4b)

## Inference

To showcase the results, we use images and videos to prove its effectiveness.

### Unet CNN  

![image](https://github.com/user-attachments/assets/b661384b-78df-4abb-b18c-4e7230e3b786)

### FPN with ResNet34  

![image](https://github.com/user-attachments/assets/de01ed13-2d44-43bd-bc80-9a54124c841a)

![image](https://github.com/user-attachments/assets/b62df75f-99e2-4060-9976-3753573b557f)

![image](https://github.com/user-attachments/assets/2671e59a-b621-45d2-b22d-d5bf6e98817f)

![image](https://github.com/user-attachments/assets/150933a6-6794-4879-8851-dacf8a2a1772)

https://github.com/user-attachments/assets/43c4bf6d-5754-47bc-aa2c-c2daaedee4ef