choidaedae / POSTECH-CSED538

POSTECH CSED538 (Deep Learning) Final Term Project: Feature Manipulation for DDPM based Change Detection

POSTECH 2023-Fall CSED538 (Deep Learning) Final Project

• Feature Manipulation on DDPM-Based Change Detection

Team

• Daehyeon Choi (POSTECH, [e-mail])
• Seunghun Oh
• Jeongho Son

Baseline Model

• DDPM-cd (2022, Under Review)

Methods

1. Feature Attention

2. FDAF (Feature Dual Alignment Fusion)

Experiments

• All our experiment results are avilable in wandb.

1. Experiments on LEVIR-CD-256 Datasets

ddpmCD-baseline-LEVIR

• Link:

ddpmCD-Attention-LEVIR

• Link:

ddpmCD-FDAF-LEVIR

• Link:

ddpmCD-FDAF-Attention-LEVIR

• Link:

2. Experiments on WHU-CD-256 Datasets

ddpmCD-baseline-WHU

• Link:

ddpmCD-Attention-WHU

• Link:

ddpmCD-FDAF-WHU

• Link:

ddpmCD-FDAF-Attention-WHU

• Link: