Number of noun categories of the object detector for v2.0

[kylemin]

Hi,

I want to produce object detections following this instruction. It works well for v1.0, but does not work for v2.0. I found that although the number of noun categories is 128, the shape of roi_heads.box_predictor.cls_score.weight of the downloaded v2.0 model (v2/sta_models/object_detector.pth) is (130, 1024). I think it should be (129, 1024).

For v1.0, the number of noun categories is 87 and the shape of roi_heads.box_predictor.cls_score.weight is (88, 1024), so it does not cause any error. I cannot load the v2.0 model and I wonder why it assumes 129 categories when there are only 128 categories in the annotations.

For v2.0, the error message looks as follows:

Skip loading parameter 'roi_heads.box_predictor.cls_score.weight' to the model due to incompatible shapes: (130, 1024) in the checkpoint but (129, 1024) in the model! You might want to double check if this is expected.
Skip loading parameter 'roi_heads.box_predictor.cls_score.bias' to the model due to incompatible shapes: (130,) in the checkpoint but (129,) in the model! You might want to double check if this is expected.
Skip loading parameter 'roi_heads.box_predictor.bbox_pred.weight' to the model due to incompatible shapes: (516, 1024) in the checkpoint but (512, 1024) in the model! You might want to double check if this is expected.
Skip loading parameter 'roi_heads.box_predictor.bbox_pred.bias' to the model due to incompatible shapes: (516,) in the checkpoint but (512,) in the model! You might want to double check if this is expected.
Some model parameters or buffers are not found in the checkpoint:
roi_heads.box_predictor.bbox_pred.{bias, weight}
roi_heads.box_predictor.cls_score.{bias, weight}