salaee/pegbis

Python implementation of "Efficient Graph-Based Image Segmentation" paper

pegbis's Issues

How to run the code?

Since there are many .py files and main makes reference to these files. How can we run the code?. It is showing undefined name reference in main for all the references made to other .py files.
Please help

Minimum spanning tree part

Why is there no minimum spanning tree part in the code ?

A Question about the code

IN main.py the 47 line if (x < width - 1) and (y < height - 2):
I think it should be " if (x < width - 1) and (y < height - 1):"

Running main.py only displays an empty plot

So I've tried running the main.py file with all the images in the data folder, but it only displays an empty plot.

I'm using python 3.6.5.

Please let me know if any additional details are required.

some functions are inconsistent with the original C++ codes.

Anyway, it works fine for tested images, but some codes are confusing

1. The sumation of filter mask in the normalization is wrong.

Original C++ version:

/* normalize mask so it integrates to one */
static void normalize(std::vector<float> &mask) {
  int len = mask.size();
  float sum = 0;
  for (int i = 1; i < len; i++) {
    sum += fabs(mask[i]);
  }
  sum = 2*sum + fabs(mask[0]);
  for (int i = 0; i < len; i++) {
    mask[i] /= sum;
  }
}```

> This version in filter.py#L30 :

```python
2 * np.sum(np.absolute(mask)) + abs(mask[0])

maybe the + before abs(mask[0]) should be -.

2. The convolve_even(src, mask) actually does not flip the output.

3. In make_fgauss: mask[i] = math.exp(-0.5 * math.pow(i / sigma, i / sigma)), the math.pow is not correctly used.

4. At main.py#L47, if (x < width - 1) and (y < height - 2):, here y < height - 1.

In my workplace, the git connection is banned, so I can't directly pull a request. 😞

create gaussian filter code likely exist some error!

mask[i] = math.exp(-0.5 * math.pow(i / sigma, i / sigma))
shouldn't mask[i] = math.exp(-0.5 * math.pow(i / sigma, 2))

segment_graph

ModuleNotFoundError: No module named 'segment_graph'
unable to install segment_graph

using google colab