Polygons are incorrectly used for BoundingBox calculation about handtracking HOT 3 CLOSED

Hey Victor,

that did the job! Great thanks!
I added sorted() to all os.listdir() functions, as they use arbitary order, just to make sure if there could be similar problems with other python / OS versions.

from handtracking.

Hi @gustavz ,

Thanks for pointing these out, this feedback is quite insightful.

• Different Results:
  The main reason the images and bounding boxes are visualized is to catch the exact sort of error you just showed. For some reason, the bounding boxes in your screen shots are not what I observe when I run same code. Below is a link to how the bounding box and polygon.mat is processed at my end.
  ALL boxes are drawn correctly.
  https://www.youtube.com/watch?v=PA0lBI_hp4w&feature=youtu.be

• Mac vs Ubuntu might be the problem:
  As to why bounding boxes are mismatched on your end, one thing I notice immediately is my tests are on a Mac (you are using Ubuntu). As you pointed out, it is possible that there are differences in how python os code processes directories in Ubuntu vs Mac (I did not factor this in .. relatively new to python). I think the fix suggested by @csaldi in issue #3 should work. i.e sorting image_path array in ascending order. My suspicion is that on a mac, the os directory walkthrough code returns files in the sorted ascending order but does not do the same on Ubuntu.

• Potential Fix:
  I just pushed a fix which sorts the image_path array. I don't have access to an ubuntu machine, but will try to do this over the next couple days.

Let me know if the fix works for you.

Thanks for the feedback!

-V.

from handtracking.

Great!
Let me know how the results from training with both datasets go!

-V.

from handtracking.