drainingsun / ybat Goto Github PK

I'm capturing data and annotating it . Here is an example image

However, when I actually upload it in the tool:

The image is rotated 90 degrees. The problem is that this causes the actual annotations to be incorrect unless I now manually rotate the image as well. Is there a way I can rotate the image inside of boobs (or not have it rotate at all?). Alternatively, the annotations could just be rotated as well. Thanks!

@drainingsun My PC shuts down because of Power Faliure.. after restarting the bboxes donot recover when i restore my browser,, serious shock... i was about to finish my work.......... i labeled 1000 images.. Help me please .......if u can recover my hard work

Hello @drainingsun
First of all thank you for this glorious tool ^^
So i was using Boobs and having no issues whatsoever but a couple of days ago i annotated some images,saved them as YOLO format and closed the tool,a couple of hours later i realized i forgot to annotate an object and re-opened all the images and their annotation files just to realize some bounding boxes on an image,espacially the small ones shifted their position on the horizontal axis a bit,they weren't in their original position,note that i wasn't able to reproduce this issue it happened only once but as you might guess this can be a real pain if one is annotating a large dataset and constantly going back to check if the previous annotations are correct =/ what may be the cause of this? Any help is greatly appreciated.

Might consider adding a more detailed description on how to use the app

Can i get the original files of the min version of js

so i tried saving a coco.json file for my dataset after i had done the annotations using ybat. I found this particular export option a bit problematic
The segmentation key needs a value with two concat arrays as shown below
"annotations":[ { "segmentation":[[ 1, 4, 1, 151, 235, 151, 235, 4 ]], . . . . }
But ybat creates this with only one array whereas it should have been
"segmentation" : [[ <some data>]]
and not this
"segmentation" : [<some data>]
see this picture to have a better understanding at what is problematic with save coco button.
https://ibb.co/wRFVYFs

Loaded class names are 'undefined' in saved txt files, although class names were successfully loaded. I get the bounding box positions but labeled as undefined

How can I do to support CoreML format?

https://github.com/drainingsun/boobs/blob/master/boobs.js#L737

The extension check seems to look for ".txt" files and loads the text after this check, however, it would be great if it was ".names" file because that's the native yolo format for class names

Can you share the source code (non-minified) for canvas.min.js, or where you got it from if you did not write it? Thanks!

now the examble flow is:

select Car object --> delete bbox label --> add new Toyota bbox label

We expect

select Car object --> edit to Toyota class object

Hye. before this, im doing text file format image lebeling for yolo object detection. And now, im using faster rcnn for object detection. But, I want to use the same dataset (from my previous yolo dataset) The problem is, my google colab notebook, does not support for text file tensorflow. so, i need to convert to xml. Is it your previous tutorial from the YT can fix my problem?

thank you so much

I find your tool very useful, however when annotating objects close to image edges, it is very easy to drag the bbox outside of the image. Adding a small feature to keep the bbox within the image would be appreciated.

The delete key on my keyboard is malfunctioning and i can not use the function，please tell me how to use it

When I use your tool. But I turn on boobs.html. I can't Bounding box . I don't know why

I have F12 but no error ???

You can fix this problem for me ???

Thanks you for contributing and reading

Can I change the class of an existing bbox?

Thanks for the tool btw, it's working very nice!

Hey there, I want to thank you for making this.

I annotated a set of images using boobs. My images include multiple appearances of the same object (in my case cars).

Here is a BOOBS-generated VOC file:

<object>
   <name>car</name>
   <pose>Unspecified</pose>
   <truncated>0</truncated>
   <occluded>0</occluded>
   <difficult>0</difficult>
   <bndbox>
   <xmin>2656</xmin>
   <ymin>607</ymin>
   <xmax>3530</xmax>
   <ymax>1879</ymax>
   </bndbox>
   <bndbox>
   <xmin>1655</xmin>
   <ymin>939</ymin>
   <xmax>2451</xmax>
   <ymax>2106</ymax>
   </bndbox>
   <bndbox>
   <xmin>665</xmin>
   <ymin>1188</ymin>
   <xmax>1506</xmax>
   <ymax>2422</ymax>
   </bndbox>
</object>

So each object gets it's own bndbox.

Is this the right VOC annoation? I thought that each object should be wrapped in it's own "object" tag. For example:

<object>
   <name>car</name>
   <pose>Unspecified</pose>
   <truncated>0</truncated>
   <occluded>0</occluded>
   <difficult>0</difficult>
   <bndbox>
   <xmin>2656</xmin>
   <ymin>607</ymin>
   <xmax>3530</xmax>
   <ymax>1879</ymax>
   </bndbox>
</object>
<object>
   <name>car</name>
   <pose>Unspecified</pose>
   <truncated>0</truncated>
   <occluded>0</occluded>
   <difficult>0</difficult>
   <bndbox>
   <xmin>1655</xmin>
   <ymin>939</ymin>
   <xmax>2451</xmax>
   <ymax>2106</ymax>
   </bndbox>
</object>
<object>
   <name>car</name>
   <pose>Unspecified</pose>
   <truncated>0</truncated>
   <occluded>0</occluded>
   <difficult>0</difficult>
   <bndbox>
   <xmin>665</xmin>
   <ymin>1188</ymin>
   <xmax>1506</xmax>
   <ymax>2422</ymax>
</bndbox>
</object>
</annotation>`