Bug

[massyzs]

line90:
loss_obj = cls_criterion(logits.unsqueeze(0), gt_obj).squeeze().mean()

It seems that the first dimension of logits is 256, and the first dimension of gt_obj is 3.

Exception has occurred: ValueError
Expected input batch_size (1) to match target batch_size (3).
File "style-splat/train.py", line 90, in training
loss_obj = cls_criterion(logits.unsqueeze(0), gt_obj).squeeze().mean()
ValueError: Expected input batch_size (1) to match target batch_size (3).

[tang-shan]

line90: loss_obj = cls_criterion(logits.unsqueeze(0), gt_obj).squeeze().mean()

It seems that the first dimension of logits is 256, and the first dimension of gt_obj is 3.

Exception has occurred: ValueError Expected input batch_size (1) to match target batch_size (3). File "style-splat/train.py", line 90, in training loss_obj = cls_criterion(logits.unsqueeze(0), gt_obj).squeeze().mean() ValueError: Expected input batch_size (1) to match target batch_size (3).

go to scene/dataset_readers.py to change line 99:
objects = Image.open(object_path) if os.path.exists(object_path) else None
to
objects = Image.open(object_path).convert('L') if os.path.exists(object_path) else None

It works for me. It seems that the segmentation maps should be a gray image, but read as a rbg image.