You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hello, I found the math of the conversion described in the README is conflicted with the example result you provided above. In the example, the first coordinate in the output is 0.8265625, but following the math, it should be (x + bbox_width / 2) / img_width) = (529 + 89 / 2) / 640 = 0.89609375. Actually, I found 0.8265625 is equal to x / img_width. When I go through the code, the implementation of the conversion equation seems to be right. Moreover, the label files provided in your FLIR_Object_Detection repo also did't follow the conversion equation shown in the README. Therefore, I am writing to find out how the current results, as well as those labels in another repo, are generated, and which one should I use, the existing labels or the conversion equations?
The text was updated successfully, but these errors were encountered:
On Fri, May 21, 2021 at 1:41 AM Yunx Dai ***@***.***> wrote:
Hello, I found the math of the conversion described in the README is
conflicted with the example result you provided above. In the example, the
first coordinate in the output is 0.8265625, but following the math, it
should be (x + bbox_width / 2) / img_width) = (529 + 89 / 2) / 640 =
0.89609375. Actually, I found 0.8265625 is equal to x / img_width. When I
go through the code, the implementation of the conversion equation seems to
be right. Moreover, the label files provided in your FLIR_Object_Detection
repo <https://github.com/joehoeller/Object-Detection-on-Thermal-Images>
also did't follow the conversion equation shown in the README. Therefore, I
am writing to find out how the current results, as well as those labels in
another repo, are generated, and which one should I use, the existing
labels or the conversion equations?
—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
<#6>, or unsubscribe
<https://github.com/notifications/unsubscribe-auth/ABHVQHGFM4AXIKAWTXMXHLLTOX6AHANCNFSM45IQWZCQ>
.
Hello, I found the math of the conversion described in the README is conflicted with the example result you provided above. In the example, the first coordinate in the output is
0.8265625
, but following the math, it should be(x + bbox_width / 2) / img_width) = (529 + 89 / 2) / 640 = 0.89609375
. Actually, I found 0.8265625 is equal tox / img_width
. When I go through the code, the implementation of the conversion equation seems to be right. Moreover, the label files provided in your FLIR_Object_Detection repo also did't follow the conversion equation shown in the README. Therefore, I am writing to find out how the current results, as well as those labels in another repo, are generated, and which one should I use, the existing labels or the conversion equations?The text was updated successfully, but these errors were encountered: