Harshit Makwana

Computer vision / Case study

Image Inpainting Using GANs

I developed a GAN-based image-inpainting model, trained it on artificially masked images and built a Streamlit interface for uploading images and viewing restorations.

Harshit Makwana

Concept illustration of a masked still-life image and a restored interpretation, not output from the project model.
AI-generated concept illustration · not a project screenshot
My role
GAN development, training & Streamlit interface
Timeline
2023–2024
Focus
Computer vision

01 / Context

Reconstructing missing image regions and making the model usable.

Missing regions interrupt an image's visual structure. I explored GAN-based inpainting to fill those regions with content that appeared realistic and coherent with the surrounding image, then made the workflow accessible through an upload-and-preview interface.

02 / Ownership

What I built

  • Developed a GAN-based model for image inpainting.
  • Trained it on images with artificially generated masks that simulated missing regions.
  • Evaluated the visual realism and coherence of the reconstructed images.
  • Built a Streamlit interface for users to upload images and inspect the inpainted results.

03 / Engineering

Technical approach

Train around the missing-region task

I used artificially created masks to simulate missing image regions in the training dataset. This gave the model examples centered on the reconstruction task rather than treating image generation as an unconstrained problem.

Evaluate the result in context

I assessed the generated regions for realism and coherence with the surrounding image. I then connected the model to Streamlit so users could upload an image and view the resulting restoration in a simple application workflow.

04 / System view

Workflow at a glance

  1. Training images

    Artificial masks simulate missing regions

  2. Model training

    GAN-based image inpainting

  3. Visual evaluation

    Realism and coherence of restored regions

  4. User interface

    Streamlit upload and result preview

05 / Delivery

Outcomes

I delivered the model workflow and a Streamlit interface for image upload and restoration review. The cover art illustrates the inpainting concept; it is not an output from my trained model or an evaluation sample.

Technologies & methods

  • Generative adversarial networks
  • Image inpainting
  • Streamlit

VehiSense Pro

I developed a vehicle-monitoring system with fine-tuned YOLOv8, tracking and event-detection logic, connected to a FastAPI web application through WebSockets.

Roadway Solutions

I developed road-inspection workflows using fine-tuned YOLO, ByteTrack, MiDaS and SAM 2, with OpenVINO experiments for inference performance.

Satellite Image Change Detection

Developed a system to detect spatial changes over time using satellite imagery and advanced image processing techniques. The solution combines deep learning models, clustering, and dimensionality reduction for environmental and urban insights.