search

cvat-ai / cvat

Annotate better with CVAT, the industry-leading data engine for machine learning. Used and trusted by teams at any scale, for data of any scale.
https://cvat.ai
MIT License
12.58k stars 3k forks source link

Exporting task to YOLO 1.1, the label file becomes 0 byte #6609

Closed prnr closed 9 months ago

prnr commented 1 year ago

My actions before raising this issue

Steps to Reproduce (for bugs)

  1. import custom coco annotation file to task 20230724112437283_coco.zip
  2. export to YOLO 1.1 format task_test-2023_08_03_03_13_44-yolo 1.1.zip
  3. All labels are blank txt file image

The imported cocojson file contains only B.box data, so I think it is not a problem that polygon is not recognized. Also, since jobs normally display B.box, there seems to be no problem until import, but I think there is a problem with YOLO 1.1 export. The imported cocojson file is a file exported while working on another platform. Could this be the cause of the problem? (some images have not been annotated)

I've tried a bunch of other coco2yolo tools, but most of them couldn't work normally (no expected results due to env issues etc), and I'm trying to convert them using cvat.

Your Environment

HawkingRadiation42 commented 1 year ago

im facing the same issue here the label files are blank, did you find any solution for it?

njoguamos commented 1 year ago

Same issue over here.

khushyrie commented 1 year ago

same issue here

njoguamos commented 1 year ago

I ended up using rectangular bounding boxes. I will consider doing object segmentation later.

khushyrie commented 1 year ago

I export the dataset into coco format first. Then convert it to yolo using roboflow

prnr commented 1 year ago

im facing the same issue here the label files are blank, did you find any solution for it?

I replaced the work with another tool. I tried many conversion methods with cvat but couldn't solve it.

njoguamos commented 1 year ago

I ended up using rectangular bounding boxes. I will consider doing object segmentation later.

I ended up doing image segmentation instead.

I exported the masks and converted them to polygons.

Turns out that segmentation is all I needed for my problem.

bsekachev commented 9 months ago
  1. When you imported COCO to CVAT, it probably was imported as polygons (check on UI how the object is recognized)
  2. YOLO format supports only bounding boxes, polygons are ignored -> files are empty

Workaround: On app.cvat.ai you may now convert polygons to rectangles:

image image

Check and save job, export to YOLO.