Home

Awesome

MSA

Code of paper “Transcending Fusion: A Multi Scale Alignment Method for Remote Sensing Image Text Retrieval” Arxiv:https://arxiv.org/pdf/2405.18959

Preparation:

  1. Setup Environment

  2. Update the data path in option/RSITMD_AMFMN.yaml to match your own data location.

  3. Modify the path for the ResNet weights on line 250 of resnet.py. Download link for ResNet weights: ResNet50: https://download.pytorch.org/models/resnet50-19c8e357.pth

  4. You will need to modify the path for /bert-base-uncased/ on line 482 of retrieval.py and line 225 of model.py. /bert-base-uncased/ can be downloaded from the Hugging Face official website.

  5. Set batch_size in option/RSITMD_AMFMN.yaml. Training epoch and learning rate can be modified on lines 480 and 481 of retrieval.py.

Execution: Run the following command: python retrieval.py

This is running convergence curve: image