The problem about the generation method of dataset
ElingBetty opened this issue · 2 comments
As seen at the line 60th in dataset.py, did you only select a query image when the number of the object pixels in this image is larger than 2 * 32 * 32 ?
I have run your code without the content about removing the line 60th in dataset.py, and I found there is a considerable drop of the performance.
We have commented in our code that we follow OSLSM (the first paper in few-shot segmentation) to implement the 60th line, and both CANet and PGNet adopt this strategy to screen out the small objects.
Of course, you can make a comparison with other methods by adding this condition following the aforementioned papers.
Also, this line mainly controls the mask sizes of the support samples, and it only slightly affects the performance. You can find more details in this issue: #11.
Thank you.