CS 838/BMI 826 Final Project
Topic: Unsupervised Segmentation by super-pixel refinement and clustering
Quick start
To evaluate our model performance, you can direcly execute:
python demo.py
| example 1 | output | example 2 | output |
|---|---|---|---|
 |
 |
 |
 |
Detailed arguments:
python demo.py
[--weights_test] # the path of model checkpoint default: (str) model/checkpoint.pth.tar
[--test_dir] # the directory of testing samples default: (str) samples/*.jpg
[--output_dir] # the directory to store output default: (str) results/
[--is_crf] # whether to use CRF or not default: (int) 0
[--img_channe] # number of input image channels default: (int) 3
[--num_channel] # number of network output channels default: (int) 100
If there is any dependency issue, please find requirement.txt to check the environment or install the required libraries by:
pip install -r requirement.txt
Training
Please find the train.ipynb to see our implementation and training details.
Evaluation
Please find the eval_IoU.ipynb to see the quantitative results. Note that
we randomly sample three results from each method and then calculate their mIoU with the ground-truth labels.