YOLO adapted version of PASCAL VOC converter.py #454
Conversation
|
Wow. Codacy/PR Quality Review seems to found issues in line that i have exactly copied from line above and changes some characters. https://app.codacy.com/app/nmanovic/cvat/pullRequest?prid=3597669 |
|
Hi @dntxos , thanks for your contribution! Our team really appreciate that. Please let us a couple of days to review and try. I will come back with a feedback shortly! |
| if ("sessionid" in httpclient.cookies): | ||
| sessionid = httpclient.cookies["sessionid"] | ||
|
|
||
| url = urlbase+"/api/v1/tasks/"+str(taskid)+"/frames/"+ str(frameid) |
There was a problem hiding this comment.
@dntxos , I general I don't think that it is a good idea to contact CVAT instance to get frames. Frames which CVAT provides using the API don't have good quality. It isn't something which should be used for training. Quality of frames should be enough to do annotation. They can be compressed and this is why we have "image quality" parameter when an annotation task is created.
There was a problem hiding this comment.
You can think that user provides you all necessary data. Thus your code will be simpler.
There was a problem hiding this comment.
Yep. This code download frames from CVAT only when a username is passed. It is an optional parameter. Since in my scenario everything works locally and i loaded with good quality (not standard 50), this seemed useful to me. I can remove from the code if you prefer. ;-)
There was a problem hiding this comment.
@dntxos, if the code is useful for you let's keep it. I will try to test your code and if I don't see any other issues I will merge. Thanks again for your contribution.
There was a problem hiding this comment.
@nmanovic, cool! Thx for your feedback. It is an honor to contribute to this amazing project. Our annotation time has considerably reduced with CVAT. 🤩👏👏👏
|
@dntxos , also need to add information about the feature to CHANGELOG.md |
utils/yolo/requirements.txt
Outdated
| @@ -0,0 +1,3 @@ | |||
| argparse>=1.1 | |||
| lxml>=3.5.0 | |||
| glog>=0.3.1s | |||
Since we use Darknet / Yolo and decided to change our annotation tool from SCALABEL to CVAT, it was necessary to write export scripts for YOLO format