Weird bounding box resize values

Describe the bug

I’m trying to resize images to half the width and height. my images are 1200x1600

Reproduction steps

from typing import Dict, Optional, Tuple, Union
import torch
from kornia import augmentation as KA
from kornia.augmentation.random_generator import random_generator as rg
from kornia.constants import Resample


class Resize(KA.RandomResizedCrop):
    def __init__(
        self, size: Tuple[int, int], resample: Union[str, int, Resample] = Resample.BILINEAR.name, *args, **kwargs
    ):
        """Resize the frame.

        Parameters
        ----------
        size : tuple of int
            size (h, w) in pixels of the cropped region
        """
        super().__init__(
            p=1,
            size=size,
            scale=(1, 1),
            ratio=(size[0] / size[1], size[0] / size[1]),
            resample=resample,
            return_transform=False,
            keepdim=True,
            *args,
            **kwargs,
        )

    def __repr__(self) -> str:
        repr = f"size={self.size}, interpolation={self.resample.name}"
        return self.__class__.__name__ + f"({repr}, {super().__repr__()})"

    def generate_parameters(self, batch_shape: torch.Size) -> Dict[str, torch.Tensor]:
        return rg.random_crop_generator(
            batch_shape[0],
            (batch_shape[-2], batch_shape[-1]),
            (batch_shape[-2], batch_shape[-1]),
            resize_to=self.size,
            same_on_batch=self.same_on_batch,
            device=self.device,
            dtype=self.dtype,
        )

KA.AugmentationSequential(            
            Resize(size=(600, 800)),
            data_keys=["input", "bbox_xyxy"],
            same_on_batch=False,
            return_transform=False,
            random_apply=True,
        )

1. Supply bounding boxes

Expected behavior

I would have expected to see half the values afterwards, yet I get this:

tensor([[458.7130, 505.5780, 482.6980, 522.5638],
        [434.2283, 478.1010, 502.6855, 533.0551],
        [455.2152, 485.5947, 480.1996, 510.5738],
        [530.6680, 421.6481, 559.1501, 464.6122],
        [383.2602, 381.6814, 425.7336, 452.1226],
        [457.7137, 466.6105, 489.1939, 540.0493],
        [544.6592, 518.5671, 577.1389, 558.5338]])

before my bounding boxes look like this

tensor([[ 918., 1012.,  966., 1046.],
        [ 869.,  957., 1006., 1067.],
        [ 911.,  972.,  961., 1022.],
        [1062.,  844., 1119.,  930.],
        [ 767.,  764.,  852.,  905.],
        [ 916.,  934.,  979., 1081.],
        [1090., 1038., 1155., 1118.]])

Environment

wget https://raw.githubusercontent.com/pytorch/pytorch/master/torch/utils/collect_env.py
# For security purposes, please check the contents of collect_env.py before running it.
python collect_env.py

- PyTorch Version (e.g., 1.0): 1.10
- OS (e.g., Linux): Linux
- How you installed PyTorch (`conda`, `pip`, source): pip
- Build command you used (if compiling from source): -
- Python version: 3.8.12
- CUDA/cuDNN version: 11.5/8
- GPU models and configuration: single gpu
- Any other relevant information:

Additional context

@shijianjian

[shijianjian]

I will update the cropping functions to be shape-agnoistic, on top of the normalised coordinates (ranged from [0, 1]). Then, I think this issue might be alleviated. Probably start next week or so.

[edgarriba]

Is it possible that the samplers fallback to center crop ? https://github.com/kornia/kornia/blob/6a970b61d222862fcc6efb483c7d4268e3ecf162/kornia/augmentation/random_generator/random_generator.py#L598