Custom trained DETR about playtorch HOT 2 CLOSED

ok, so it looks like I need to index from 1 and the the specfile looks like.

"unpack": {
    "type": "bounding_boxes",
    "key": "boundingBoxes",
    "dtype": "float",
    "probabilityThreshold": "$probabilityThreshold",
    "classes": [
       "N/A",
      "test class"
    ]
  }

Fine it works, but it would be nice to have it stated explicitly. btw., are you looking for tutorials? I could make one for retraining detr

from playtorch.

Thanks for flagging this, @batrlatom!!

are you looking for tutorials? I could make one for retraining detr

Yes, contributions to PyTorch Live are very much welcome, including new tutorials! Feel free to submit a DETR tutorial as PR!!

FYI, the classes in the example are based on the CoCo dataset, similar to what is used in DETR panoptic in this notebook

from playtorch.