prevent duration = 0.0 and divding by zero

[sitifukujin]

It is small bug. On my windows, it occurs dividing zero 100%. It might occur on linux.

[yuyu2172]

LGTM
Thank you for your contribution!

[Hakuyume]

How about showing inf for speed when duration is zero?

[sitifukujin]

Hello again, Hakuyume-san,
Would you want to see stack trace?

Downloading from http://host.robots.ox.ac.uk/pascal/VOC/voc2007/VOCtest_06-Nov-2007.tar...
Traceback (most recent call last):
  File "D:\DeepLearning\Github\sitifukujin_chainercv\examples\detection\visualize_models.py", line 53, in <module>
    main()
  File "D:\DeepLearning\Github\sitifukujin_chainercv\examples\detection\visualize_models.py", line 18, in main
    dataset = VOCDetectionDataset(year='2007', split='test')
  File "D:\DeepLearning\Github\sitifukujin_chainercv\chainercv\datasets\voc\voc_detection_dataset.py", line 72, in __init__
    data_dir = voc_utils.get_voc(year, split)
  File "D:\DeepLearning\Github\sitifukujin_chainercv\chainercv\datasets\voc\voc_utils.py", line 34, in get_voc
    download_file_path = utils.cached_download(urls[key])
  File "D:\DeepLearning\Github\sitifukujin_chainercv\chainercv\utils\download.py", line 72, in cached_download
    request.urlretrieve(url, temp_path, _reporthook)
  File "C:\Anaconda3\lib\urllib\request.py", line 284, in urlretrieve
    reporthook(blocknum, bs, size)
  File "D:\DeepLearning\Github\sitifukujin_chainercv\chainercv\utils\download.py", line 27, in _reporthook
    speed = int(progress_size / (1024 * duration))
ZeroDivisionError: float division by zero
Press any key to continue . . .

[Hakuyume]

@sitifukujin I think using small value for duration is not a good solution because it shows nonsense speed. How about showing 'inf KB/s' when the duration is zero. User can know that the speed can't be computed now.

[Hakuyume]

How about using try and except to catch ZeroDivisionError. I think this is better because we can remove a magic value 1e-4.

[sitifukujin]

@Hakuyume
Line this?

Downloading from http://host.robots.ox.ac.uk/pascal/VOC/voc2007/VOCtest_06-Nov-2007.tar...

...0, 0.0078125 MB, inf KB/s, 0.0 seconds passed
...0, 0.015625 MB, 59 KB/s, 0.26875853538513184 seconds passed
...0, 0.0234375 MB, 89 KB/s, 0.26875853538513184 seconds passed

[Hakuyume]

@sitifukujin Yes. I think this output is better. I added a comment. Please check it.

[sitifukujin]

Please check it again.

[Hakuyume]

LGTM

[sitifukujin]

Thanks, Yuyu-san and Hakuyume-san. See you.