search

ryouchinsa / Rectlabel-support

RectLabel is an offline image annotation tool for object detection and segmentation.
https://rectlabel.com
507 stars 73 forks source link

Bag with orientation and coordinates #167

Closed GLKonstantin closed 4 years ago

GLKonstantin commented 4 years ago

Программа отображает ориентацию изображения согласно метки exif файла, а вот размеченные координаты сохраняет согалсно ориентации оригинала, таким образом в обучение могут попасть неверно размеченные данные что искажают обучение. Такое может произойти если вы вручную повернули изображение перед тем как обработать его в RectLabel. Эта статья описывает проблему и решение: https://habr.com/ru/post/471112/

The program loads the image orientation according to the exif label in the file, and saves the coordinates relative to the original image. As a result, incorrectly marked data ends up in machine learning. This article describes the problem and solution: https://habr.com/ru/post/471112/

ryouchinsa commented 4 years ago

Thanks for writing the issue.

Which neural network tool are you using to train your object detection model?

sizeMismatch error on detectron2 https://github.com/facebookresearch/detectron2/issues/863

RectLabel and macOS Viewer show images with Exif orientation flags in the same way. According to the Exif orientation flag, each image is rotated and shown in the front orientation. So that the annotation file on RectLabel assumes that the image is rotated and shown in the front orientation. https://rectlabel.com/help#open_both_folders

EXIF Orientation-flag example images https://github.com/recurser/exif-orientation-examples

EXIF Orientation Handling Is a Ghetto https://www.daveperrett.com/articles/2012/07/28/exif-orientation-handling-is-a-ghetto/

GLKonstantin commented 4 years ago

Sorry friend. That's my fault. I cut the image according to the * .json coordinates for create ml and got the wrong places in the coordinates. I programmatically opened the image and did not watch exif.   I am really looking forward to the ipad application and will definitely buy it.            

Суббота, 30 мая 2020, 4:57 +03:00 от Ryo Kawamura notifications@github.com:     Thanks for writing the issue. Which neural network tool are you using to train your object detection model? sizeMismatch error on detectron2 facebookresearch/detectron2#863 RectLabel and macOS Viewer show images with Exif orientation flags in the same way. According to the Exif orientation flag, each image is rotated and shown in the front orientation. So that the annotation file on RectLabel assumes that the image is rotated and shown in the front orientation. https://rectlabel.com/help#open_both_folders EXIF Orientation-flag example images https://github.com/recurser/exif-orientation-examples EXIF Orientation Handling Is a Ghetto https://www.daveperrett.com/articles/2012/07/28/exif-orientation-handling-is-a-ghetto/ — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub , or unsubscribe .     С уважением, Левицкий Константин Аркадьевич.

т.моб. 8(903)751-87-01 89037518701@mail.ru  

ryouchinsa commented 4 years ago

Currently If you need our support to fix this problem, please let us know.

ryouchinsa commented 1 year ago

Currently RectLabel has "Remove EXIF orientation flags" feature on File menu. After opening images, we recommend to remove EXIF orientation flags. If necessary, you can rotate the image by 90 degrees using "Rotate the image right/left" feature on Edit menu.