acl21 / image_bbox_slicer Goto Github PK

I am trying to use the image_bbox_slicer in that it fits exactly my use case. My problem is that it keeps through exceptions like below no matter what filenames or paths I used. I looked at your code a little bit thinking that maybe it is my file names which are like "Scenario_10_3_1_standing_by_vehicle_150ft-0058.jpg". I did not see anything in your code that would have a problem with the file name. Do you have any suggestions on what I am doing wrong?

Exception: Each image in E:/sd_data_collection/sd_11_15_19/ripped_frames/tile_test/test_src/images must have its corresponding XML file in E:/sd_data_collection/sd_11_15_19/ripped_frames/tile_test/test_src/annotations with the same file name.

Hi,

This is a fab tool but unfortunately, it places the source filename in the XML not the tile filename

<filename>**ORIGINAL_FILENAME_WITHOUT_FILETYPE**</filename>
<path>**ORIGINAL_FILEPATH_WITHOUT_FILETYPE**</path>
<source>

When using the scripts from the Tensorflow object detection API tutorial, this causes creation of TF records to fail as it looks for files, which arent there
https://tensorflow-object-detection-api-tutorial.readthedocs.io/en/latest/training.html#preparing-the-dataset

I will work on a PR to address it but wanted to raise this as an issue

After passing through the object detection model how to get the original image back with predicted coordinated for each object?

Hi!
First of all i just want to say what a great tool this is!

I'm wondering if there is any possibility to merge sliced annotations back to it's original image?

Let's say i slice an image, train a model for object detection, predict on a new image sliced as the training data. I would then want to merge the predicted bounding boxes with the non-sliced image.

Any idea how i would go about?

Cheers!

Hi,

I am getting the above error due to the image size being (0,0). I have a JPEG image in the src/image folder but for some reason the image slicer is not reading it properly. How do I get the slicer to properly read this?

Thanks!

I would to have the original name of the image/annotation following a index to distinguish different files. How can I change the code?

Congratulations

Hey, validate_file_names function does not work properly and splits the filename multiple times when there are multiple chars of '.' in the filename other than extension's. This can be solved by replacing split method with os.splitext on those two lines:

+imgs = [os.path.splitext(x.split(separator)[-1])[0] for x in imgs]
+anns = [os.path.splitext(x.split(separator)[-1])[0] for x in anns]
-imgs = [x.split(separator)[-1].split('.')[-2] for x in imgs]
-anns = [x.split(separator)[-1].split('.')[-2] for x in anns]

For example filename = mert1.atagul2.png generated img name = atagul2

I have used the tool to slice my image and its bbox annotations(keep partial labels and ignore empty tiles). Some bboxes coordinates are -1, and I wonder what made this outcomes.