search

prabindh / euclid

Euclid object labeller for frictionless object detection training purposes in Machine learning frameworks (KITTI, YOLO)
45 stars 20 forks source link

Label file doesn't contain the correct classes #1

Closed DarylWM closed 7 years ago

DarylWM commented 7 years ago

I labelled several objects as class 1, another object as class 2, and another as class 3. However, the saved label file labelled the first object as class 0, and all the others as class 1.

image

And the entries in the label file: 0 0.8408724 0.5775401 0.0775444 0.4866310 1 0.7257674 0.6898396 0.0686591 0.5935829 1 0.4204362 0.5561497 0.0201939 0.2192513 1 0.3994346 0.5561497 0.0218094 0.2299465 1 0.4697092 0.5307487 0.0185784 0.1844920 1 0.6466074 0.7286096 0.0751212 0.2967914 1 0.6021809 0.3850267 0.0395800 0.7005348

prabindh commented 7 years ago

Will get back with an update shortly.

prabindh commented 7 years ago

I just checked, and it does not behave like what you are seeing. It is saving the correct class IDs (0 based index). Please use "git pull origin master" to get the latest version. Also, in the latest version, I have added a line in Line# 373 (commented out, you can uncomment), that will show (via Alert popup) the Class Label that is getting saved each time.

DarylWM commented 7 years ago

It's all good - I didn't get that I needed to change to 'KITTI' mode on line 116.

Thanks for writing this utility. It's just what I needed to create a new DetectNet data set.

prabindh commented 7 years ago

There was a critical bug fixed in parsing of already existing labels in last update. In addition, an enhancement to be able to use the Keyboard "x" to draw bounding box instead of mouse click, has been added for faster processing of databases. Please check latest commit (1/18/2017).