Project based on the following proposal: Proposal PDF
Drive link for daily updates and revised timeline: Drive Link
Kaggle Notebook: Notebook
Download Model and Network architecture from the drive link: Model and Architecture
import torch
from PartialConvArch256 import *
model = PartialConvUNet()
model = torch.load('ImageInpainting600k.pt', map_location=torch.device('cpu')) Week 1: 15 – 21 December
Learn basics of pytorch and understand how to implement basic features
As a learning practice, follow a flower classification tutorial to learn how to implement CNN in pytorch
Learn math behind CNNs and Partial convolution networks
Learn basics of open Cv and how to create simple bitmasks. Learn how to process base64 bitmasks to communicate with the website and the model processor
Set up a basic website using bootstrap to allow users to upload their images
Week 2: 22 – 28 December – until mid-evaluation
Start on data collection and organization.
Download and set up the ImageNet dataset. Learn about epochs in Pytorch
Use OpenCV to generate random bitmasks for the network
Start working on the main architecture of the model in Pytorch following the research paper.
Start training the model on a smaller scale to test for bugs and confirm main functionality
Week 3: 29 Dec – 4 Jan
Start working on larger scale and testing
If the model is not returning satisfactory results, try diffusion models
Learn about diffusion models and how to implement them in Pytorch if PCN doesn’t work
Start working on more functionality in website. Such as more selection tools (rect, circle, polygon)
As a secondary goal, work on semantic segmentation models (SSM)for more selectivity in the image
Find and sanitize the dataset for SSMs
Start training on semantic segmentation model
Week 4: 5 – 11 Jan
Work on integration of semantic segmentation model with the existing website so users can directly select unwanted objects and remove them completely
# todo: prupose of project, networks and dataset used.
Final touchups
12 Jan – Final Evaluation
Work on documentation and cleanup