FireGAN

FireGAN

< Go back

A Generative Adversarial Network (GAN), based on the CycleGAN architecture, that is trained to generate fire in images of buildings. (Don't ask why). The model was implemented with PyTorch, Served with a Flask backend to be used from a React web app.

  • STACK Python, PyTorch, Flask, TypeScript, React
Trained on a dataset created of images collected from the internet, containing a combination of buildings, fires, natural accidents, forest fires, building fires, etc. Trained for 200 epochs each, with and without using the identity loss. ~20 hours on an NVIDIA Tesla V100.

Selected Outputs

Output 1
Output 2
Output 3
Output 4
Output 5
Output 6
Output 7
Output 8
Output 9
Output 10
Output 11
Output 12
Output 13
Output 14
Contact
220 smtp.example.com ESMTP Postfix
HELO client.example.com
250 smtp.example.com
MAIL FROM: <me@mydomain.com>
250 Ok
RCPT TO: <hi@zaeem.dev>
250 Ok
DATA
354 End data with <CR><LF>.<CR><LF>
From: "Me" <me@mydomain.com>
To: "Jasir" <hi@zaeem.dev>
Date: $Tue, 16 Sep 2025 17:17:42 GMT
Subject: Complaint Mail

Hi Jasir,
Your site gave me a virus. I am very upset.
Yours truly,
Me
.
250 Ok: queued as 12345
QUIT
221 Bye

Say Hi!

You can reach me at mailto:hi@zaeem.dev

Site best viewed at 1920x1080 in Your Browser