SpectraDiff: Enhancing the Realism and Physical Consistency of Infrared Image Translation with Semantic Guidance

Status

• SpectraDiff Model Pipeline
• Sample Train/Inference
• Metrics (PSNR, SSIM, FID, LPIPS, DISTS)

Usage

Environment

pip install -r requirements.txt

Pre-trained Model & Configuration

Dataset	Safetensors & Configuration
FLIR	Google Drive
FMB	Google Drive
MFNet	Google Drive
RANUS	Google Drive
IDD-AW	Google Drive

Data Prepare

For segmentation: https://github.com/IDEA-Research/Grounded-Segment-Anything

Dataset	URL
FLIR	URL
FMB	URL
MFNet	URL
RANUS	URL
IDD-AW	URL

Set Configuration

The config_base.yaml file contains important settings for both training and testing.

Make sure to review and modify this file if you’re using custom data or want to alter the network structure.

Please refer to the provided configuration file and update the data root in the data section.

Training

1. Navigate to the UNET/trainer folder
2. Run one of the following commands:

python train.py

or

accelerate launch train.py

Inference

1. Navigate to the UNET/tester folder
2. Modify the configuration path and safetensors path in test.py
3. Run the following command:

python test.py