Object Detection with images of different sizes

Trying to do Object detection on a batch of images with different sizes.

I have 10 images in my batch:

from transformers import pipeline
from PIL import Image

# If doing CPU inference, set device="cpu" instead.
obj_detector = pipeline("object-detection", model="facebook/detr-resnet-50", device="cuda:0", do_pad=False)
outputs = obj_detector([Image.fromarray(image_array) for image_array in batch], top_k=1, batch_size=10)

batch is a list of image numpy arrays with each image having a different height and width.

But run into this issue

RuntimeError: The expanded size of the tensor (1066) must match the existing size (800) at 
non-singleton dimension 1.  Target sizes: [1066, 1066].  Tensor sizes: [1066, 800]