AdaptFormer, uniquely designed to adaptively interpret hierarchical semantics. Instead of a one-size-fits-all approach, it strategizes differently across three semantic depths: employing straightforward operations for shallow semantics, assimilating spatial data for medium semantics to emphasize detailed interregional changes, and integrating cascaded depthwise attention for in-depth semantics, focusing on high-level representations

Here is how to use this model to classify an image:

python

from transformers import AutoImageProcessor, AutoModel
from PIL import Image
import requests

image_processor = AutoImageProcessor.from_pretrained("deepang/adaptformer-LEVIR-CD")
model = AutoModel.from_pretrained("deepang/adaptformer-LEVIR-CD")

image_A = Image.open(requests.get('https://raw.githubusercontent.com/aigzhusmart/AdaptFormer/main/figures/test_2_1_A.png', stream=True).raw)
image_B = Image.open(requests.get('https://raw.githubusercontent.com/aigzhusmart/AdaptFormer/main/figures/test_2_1_B.png', stream=True).raw)
label = Image.open(requests.get('https://raw.githubusercontent.com/aigzhusmart/AdaptFormer/main/figures/test_2_1_label.png', stream=True).raw)


inputs = preprocessor(images=(image_A, image_B), return_tensors="pt")
outputs = adaptfromer_model(**inputs)
logits = outputs.logits # shape (batch_size, num_labels, height, width)
pred = logits.argmax(dim=1)[0]

License

The license for this model can be found here.

BibTeX entry and citation info

bibtex

@article{huang2024adaptformer,
  title={AdaptFormer: An Adaptive Hierarchical Semantic Approach for Change Detection on Remote Sensing Images},
  author={Huang, Teng and Hong, Yile and Pang, Yan and Liang, Jiaming and Hong, Jie and Huang, Lin and Zhang, Yuan and Jia, Yan and Savi, Patrizia},
  journal={IEEE Transactions on Instrumentation and Measurement},
  year={2024},
  publisher={IEEE}
}