Astro_CBAM_Unet
Description
1.Testing dataset; 2. Checkpoint; 3.Astro_CBAM_Unet model
Files
Steps to reproduce
The results presented in this study can be reproduced by following these systematic steps:Data Preparation:Acquire the HUMG UAV Dataset (0.1 m resolution) and/or the WHU Building Dataset (0.3 m resolution). Tile the high-resolution imagery into 512 \times 512 pixel patches. Apply on-the-fly data augmentation, including horizontal/vertical flips (50% probability), random color jittering (0.2 factor), and stochastic rotations within 15 degree to enhance model generalization. Model Configuration:Implement the Astro-ResCBAM-Unet architecture using the PyTorch framework. Configure the encoder with four down-sampling blocks integrated with Convolutional Block Attention Modules (CBAM) for adaptive feature refinement. Set up the bottleneck using four sequential ResConvBlocks with increasing dilation rates (2, 4, 8, 16) to expand the receptive field. Link the encoder and decoder via residual skip connections to stabilize gradient flow. Training Protocol: Employ the AdamW optimizer with an initial learning rate of 0.001. Apply a Cosine Annealing Scheduler to dynamically adjust the learning rate over a maximum of 50 epochs. Execute training with a mini-batch size of 4 images on an NVIDIA Tesla T4 GPU (or equivalent). Evaluation: Assess the final segmentation masks using pixel-level metrics: Intersection over Union (IoU), F1-score, Precision, and Recall.