Receptacle M-RCNN label set

[Axe--]

Hey, for the pretrained receptacle Mask RCNN, where is the corresponding label set (classes) that contains the 32 categories?

The alfworld_receptacles variable contains 38 classes, and upon visualizing the bboxes, the labels predicted are incorrect (mismatch).

Could you assist with the same, Thanks!

def load_pretrained_model(path, device, which_type):
    if which_type == 'obj':
        categories = len(object_detector_objs)
    elif which_type =='recep':
        categories = 32
    mask_rcnn = get_model_instance_segmentation(categories+1)
    #pickle.dump(torch.load(path, map_location=device), open("loaded.p", "wb"))
    mask_rcnn.load_state_dict(torch.load(path, map_location=device))
    return mask_rcnn

[soyeonm]

Hello, sorry for my late reply. I will reply back in a few days!

[joeyy5588]

I think you can find this one in https://github.com/alfworld/alfworld/blob/master/alfworld/gen/constants.py#L1206 since they are using the pretrained detector from this repo.

[ByZ0e]

Hi! Having the same confusion. It seems that only 16 receptacle objects are segmented. What are they and what are the correspondences between them and the visualization color palette? Thanks :)