Laboratory Work

Overview

This work was conducted based on the following paper: Active Domain Adaptation via Clustering Uncertainty-Weighted Embeddings (ICCV 2021)

Objective

The main goal of this study was to evaluate whether the proposed approach is applicable to the segmentation task.

Implementation Details

The core implementation is located in the src directory. A complete description of all parameters is provided in the run.py file.

Pretrained UNet Model

A pretrained UNet model can be downloaded from the following link:
UNet Model

Running the Code

Step 1: Setting Up the Environment

To create a virtual environment and install dependencies, run the following commands:

python3 -m venv CLUE 
source CLUE/bin/activate 
pip install -r requirements.txt

Step 2: Activating the Environment

Before running the code, activate the virtual environment:

source ./CLUE/bin/activate

Step 3: Running the Script

Execute the following command to run the code with default parameters:

python3 run.py --train False --num_clusters 20 --clue_softmax_t 0.1 \
--adapt_num_epochs 10 --device cuda:0 --uncertainty CrossEntropy \
--kernel_size 5 --stride 2 --target_size 4096

Notes

• Ensure that the correct CUDA device is specified in the --device parameter.
• Modify the parameters as needed to adapt to different segmentation tasks.