microsoft/MaskFlownet

bugfix for kitti

chengrongliang opened this issue · 3 comments

Hi Simon,
some bugfix as follow.

  1. reader/kitti.py: line44 and line 99
    samples = None ->samples =-1;
    because samples = 32 if args.debug else -1 in main.py line194;
  2. reader/kitti.py: line98
    num_files = (len(os.listdir(path_testing)) - 1) // 2 -> num_files = len(os.listdir(path_testing)) // 2;
  3. reader/kitti.py: line105 & line106
    img0 = cv2.resize(img0, resize)
    img1 = cv2.resize(img1, resize)
    ->
    adj_resize = (resize[1], resize[0])
    img0 = cv2.resize(img0, adj_resize)
    img1 = cv2.resize(img1, adj_resize)
    because cv2.resize() should be dim(width, height) -> dim(cols, rows)

Hi Rongliang, thanks a lot for bugfixing!

We will fix them in the repo soon.

As I see you haven't updated this fix in the repo, could you please explain why:
reader/kitti.py: line44 and line 99
samples = None ->samples =-1;
because samples = 32 if args.debug else -1 in main.py line194;

Hi @chengrongliang ,

Thanks for sharing this bug.

I'm wondering if you can get the evaluation results of Kitti 2012/2015 as they reported in the table (e.g., 3.25, 0.11 | 9.14, 0.18 for dbbSep30) or not.

I have modified the codes and fixed the bugs, however, the calculated AEPE and F-1 are not matching.
Hope to hear from you soon.